Key card positioning within a vehicle

Abstract

A system for locating an ultra-wideband (UWB) device includes a UWB transceiver that identifies and extracts features of at least one channel impulse response (CIR), and a special-purpose processor that applies a machine learning classification process to the extracted CIR features to locate the UWB device in a vehicle.

Description

Technical Field

[0001] This disclosure generally relates to ultra-wideband (UWB) ranging technology, and more specifically, to implementing a combination of machine learning and UWB technology to determine the location of a key card inside a vehicle. Background Technology

[0002] Key cards are commonly used to provide keyless entry for modern vehicles such as cars, trucks, and boats. For example, a key card can remotely transmit electronic signals to command the vehicle to lock or unlock doors, activate the ignition to start the engine, etc., when a user approaches.

[0003] Ultra-wideband (UWB) technology is implemented in wireless electronic communication transceivers because it offers superior positioning accuracy and high data throughput compared to other wireless technologies, especially those that include Internet of Things (IoT) devices. One application involves implementing a UWB computer chip in a key card, smartphone, or other remotely mobile electronic device for remote digital access and control of the vehicle's computer processor via one or more antenna anchors installed in the car, thereby locking or unlocking doors or using other remote functions. For example, when a key card is in the driver's possession, the vehicle can detect activity such as the driver approaching the vehicle or pulling a door handle and begin searching for the key card outside the vehicle cabin. Once the key card is authenticated, the door unlocks automatically.

[0004] A key feature of modern remote keyless vehicles is that they prevent the vehicle from locking itself when the key card is inside the vehicle. However, locating the UWB key card inside the vehicle cabin presents a challenge. Specifically, UWB radios perform distance estimation algorithms, such as ToA, Time of Flight (ToF), Time Difference of Arrival (TDoA), and Angle of Arrival (AoA), which can be prone to errors and miscalculations caused by environmental conditions or by people, animals, luggage, or other objects coexisting with the key card inside the vehicle cabin. Summary of the Invention

[0005] According to one aspect of the present invention, a system for locating an ultra-wideband (UWB) device is provided, comprising: a UWB transceiver that identifies and extracts features of at least one channel impulse response (CIR); and a dedicated processor that applies a machine learning classification process to the extracted CIR features to locate the UWB device in a vehicle.

[0006] According to one or more embodiments, the UWB device is a key card.

[0007] According to one or more embodiments, the UWB transceiver includes a CIR feature processor that extracts features from the at least one CIR for analysis and subsequent processing via the machine learning classification process.

[0008] According to one or more embodiments, a machine learning computer distinguishes reflections in the at least one desired CIR for identifying a first path in the dense multipath environment of the vehicle.

[0009] According to one or more embodiments, the CIR feature processor further normalizes, filters, and / or labels the at least one CIR.

[0010] According to one or more embodiments, the UWB transceiver includes:

[0011] antenna;

[0012] A ranging processor, which generates the at least one CIR based on a UWB signal received from the UWB device by the antenna; and

[0013] A signal detector that detects and quantizes the UWB signal received by the UWB transceiver.

[0014] According to one or more embodiments, the UWB transceiver includes a housing, and the dedicated processor is housed within the housing.

[0015] According to one or more embodiments, the dedicated processor applies a single classifier from the machine learning classification process to the extracted CIR features.

[0016] According to one or more embodiments, the dedicated processor is configured and arranged to receive and process distance data about the UWB device for training the machine learning classification process to distinguish the location of the UWB device inside the vehicle.

[0017] According to a second aspect of the present invention, an ultra-wideband (UWB) wireless communication system is provided, comprising:

[0018] Key card, the key card outputs a UWB signal;

[0019] At least one anchor device, which participates in a ranging sequence together with the key card, the ranging sequence including generating at least one channel impulse response (CIR) based on the UWB signal and extracting features of the at least one CIR; and

[0020] A dedicated processor applies a machine learning classification process to the extracted CIR features to locate the key card.

[0021] According to one or more embodiments, at least one anchor includes a plurality of anchor devices, each anchor device storing and processing a classifier for the machine learning classification process regarding the CIR features.

[0022] According to one or more embodiments, the machine learning classification process selects the CIR features to train the classifier.

[0023] According to one or more embodiments, the machine learning classification process executes a training algorithm and returns a model for the classifier as output, the classifier being used to classify data with the CIR features.

[0024] According to one or more embodiments, a machine learning computer distinguishes reflections in the at least one desired CIR for identifying a first path in the dense multipath environment of the vehicle.

[0025] According to one or more embodiments, the at least one anchoring device includes:

[0026] antenna;

[0027] A ranging processor that generates the at least one CIR based on a UWB signal received by the antenna;

[0028] A signal detector that detects and quantizes the UWB signal received by the UWB transceiver; and

[0029] A CIR feature processor extracts the features of the at least one CIR.

[0030] According to one or more embodiments, the at least one anchor device includes a housing, and the dedicated processor is housed within the housing.

[0031] According to another embodiment of the present invention, a method for locating an ultra-wideband (UWB) device is provided, comprising:

[0032] The UWB device outputs a UWB signal to the UWB transceiver.

[0033] The UWB transceiver extracts features of at least one channel impulse response (CIR) of the UWB signal; and

[0034] A dedicated processor applies a machine learning classification process to the extracted CIR features to locate the UWB device in the vehicle.

[0035] According to one or more embodiments, it further includes:

[0036] Processing the CIR to distinguish reflections in the CIR of the UWB signal; and

[0037] Identify the first path in the dense multipath environment of the UWB device.

[0038] According to one or more embodiments, processing the CIR further includes generating data for the machine learning classification process, wherein the machine learning classification process executes a training algorithm to return a model of a classifier for classifying the CIR features.

[0039] According to one or more embodiments, the machine learning classification process estimates the location of the UWB device in the vehicle cabin. Attached Figure Description

[0040] This invention is illustrated by way of example and is not limited to the accompanying drawings, in which similar reference numerals indicate similar elements. The elements in the drawings are shown for simplicity and clarity, and these elements are not necessarily drawn to scale.

[0041] Figure 1 This is a graph showing the channel impulse response (CIR) generated in an environment where the concepts of the present invention are practiced.

[0042] Figure 2 This is a graph illustrating the channel impulse response generated in a dense multipath environment in an embodiment of the present invention.

[0043] Figure 3 This is a schematic view of an ultra-wideband (UWB) ranging system implemented in a vehicle according to some embodiments.

[0044] Figure 4 This is a block diagram of a UWB communication system according to some embodiments.

[0045] Figure 5 This is a flowchart illustrating a method for locating a key card inside a vehicle cabin according to some embodiments.

[0046] Figure 6 This is a flowchart illustrating a method for training a machine learning classification algorithm to locate a key card, according to some embodiments.

[0047] Figure 7 It shows the basis Figure 6 The machine learning technique selects the features of CIR for preprocessing as a graph. Detailed Implementation

[0048] UWB radio devices typically require CIR signal measurements to estimate when a signal will arrive at the receiver. The measured CIR is determined from a cross-correlation function calculated according to the IEEE 802.15.4a standard, which is a comparison between the received pulse sequence signal on the packet preamble and a reference or expected pulse. Once obtained, the CIR is used to estimate the timing of the first path between the transmitting and receiving UWB radio devices. Figure 1 and 2 The example shown in the graph illustrates a CIR for an environment with few reflections and a second path that is clearly distinguishable from the first path and may also identify additional reflections.

[0049] However, relevant environmental conditions for use by UWB radio devices include, for example, a "dense multipath" environment when attempting to locate a key card inside a vehicle cabin, where the vehicle detects activity such as a user attempting to open a door by pulling a door handle, and in response, the vehicle begins searching for the key card. (Reference) Figure 2 In dense multipath environments, it can be difficult to distinguish reflections included in CIR. Specifically, traditional search and distance estimation algorithms or localization algorithms are often insufficient for detecting direct-path signals in the presence of dense multipath conditions when dealing with errors in distance and location estimation. This is because these algorithms focus on acquiring distance information from UWB devices and must account for signal reflections (e.g., radio frequency (RF)) and scattering along the line-of-sight (LOS) path between the UWB transmitter and receiver, which can introduce interference and thus signal distortion at the receiver antenna. Multipath produces incorrect time-of-arrival (ToA) estimates, leading to erroneous distance and location estimates. Nevertheless, CIR signal data analysis can be performed, but considering dense multipath CIR can affect the accuracy of distance estimation calculations between transmitting and receiving UWB radio devices. Therefore, some “dense multipath” environments are highly unfavorable for localization using UWB devices.

[0050] In short, embodiments of the present invention combine computer-executed machine learning processes with measured CIR features, rather than relying solely on distance information to estimate the location of a keycard or associated UWB device within a vehicle, even in the presence of dense multipath environments. In a preferred embodiment, a machine learning technique is provided that includes a classification algorithm for estimating the location of a keycard within the vehicle cabin or other enclosed areas susceptible to dense multipath conditions. The machine learning method identifies characteristics or features of the CIR, such as… Figure 1 and 2As shown, this identifies the CIR as unique relative to each location within the vehicle cabin. Unlike the complexity and substantial processing requirements of deep learning techniques, machine learning recognizes that it's impossible to estimate the exact coordinates of the keycard; instead, it relies on the characteristics of the signal to identify the location, even though no xy coordinates are determined except for identifying the presence of objects, people, etc., near the keycard. For example, a machine learning algorithm is executed that learns from specific defined CIR features and continues to learn or train a classification algorithm based on the learned CIR feature data. Similarly, because the computational performance requirements are less than those of deep learning algorithms, a simple machine learning model can be executed at the UWB anchor. The machine learning algorithm allows for the selection of CIR features to train the classifier. In this way, a trade-off between the computational complexity of identifying features and the required accuracy to reduce computational requirements is possible. Therefore, although various obstacles, such as human bodies, may be present in the vehicle, the classification algorithm can be trained at execution to identify whether the keycard is on a passenger seat, driver's seat, floor, trunk, engine, etc. Another feature is that each UWB anchor in the vehicle requires a single classifier, and therefore, there are fewer anchors and less complexity. Another advantage over deep learning algorithms is that machine learning algorithms are implemented by processing CIR features or characteristics used to train the algorithm.

[0051] In some embodiments, a positioning system including a UWB transceiver such as a key card, an anchor, and a machine language computer (not shown) can exchange data to process a combination of CIR data from the CIR and data derived from the machine language, for example, to derive an accurate estimate of the distance between the UWB transceiver and the anchor. The distance data can be used to train the machine language and is derived from the CIR. The decision to process the positioning determination can be based on two inputs: the CIR and the CIR data derived from the machine language. For example, a "hybrid" system could thus process a combination of CIR data from the CIR and data derived from the machine language, as well as distance data from one or more anchors. The system can process both inputs to eliminate any ambiguity in CIR-based distance-only determinations.

[0052] Figure 3 This is a schematic view of an ultra-wideband (UWB) ranging system implemented in a vehicle according to some embodiments.

[0053] The key card 102 ranging starter device is located inside the vehicle 12, such as in the cabin, cockpit, or galley of a car, truck, aircraft, or ship, but not limited to these. The key card 102 is typically portable and includes a transmitter, activated, for example by a button, voice, or other activation mechanism, for sending commands to one or more in-vehicle transceivers 104A, 104B (typically 104) to perform well-known functions, such as communicating with various sensors 15 and associated mechanisms and an in-vehicle computer for unlocking doors, starting the vehicle engine, etc. The transceiver 104 can be located, for example, inside or outside the vehicle 12, and can include an external antenna at various locations within the vehicle, such as door handles, mirrors, trunk, bumpers, roof, dashboard, etc. As described herein, a system including the key card 102, the in-vehicle transceiver 104, and the machine learning system 148 can locate the key card 102 inside the vehicle 12 even in dense multipath environments, and can therefore be used in applications related to remote keyless entry, remote engine start, remote door locking / unlocking, etc.

[0054] For example, when the vehicle is triggered by a touch detected by sensor 15, a signal is transmitted from the antenna to the key, allowing the vehicle to unlock automatically in a subsequent authentication scheme. In some embodiments, transceiver 104 is a UWB anchor that exchanges data with key card 102 in a predetermined UWB frequency band that provides a considerable communication range, such as up to several hundred feet or more. In some embodiments, UWB transceiver 104, etc., conforms to the IEEE 802.15.4 system and typically transmits pulses at 500 MHz or 1 GHz, but is not limited thereto.

[0055] Vehicle 12 may also include a body control module (BCM) 106 or related electronic control units for various sensors 15 or onboard computers, for controlling electronic components in the vehicle body such as door locks, ignition systems, power windows, lights, and air conditioning units. Vehicle 12 may include a data bus (not shown) and / or other networking devices for allowing data exchange between BCM 106, sensors 15, onboard computers (not shown), action relays (not shown), and key card 102.

[0056] Figure 4 This is a block diagram of a UWB communication system 10 according to some embodiments. In some embodiments, the UWB communication system 10 conforms to the IEEE 802.15.4 technical standard, which is incorporated herein by reference in its entirety. As shown, a key card 102 and an onboard UWB transceiver 104 exchange data through one or more transmission channels 110, collectively referred to as a communication path or link. Embodiments of the UWB communication system 10 may include... Figure 3Implemented in key card 102, UWB transceiver 104, BCM 106 and / or sensor 15.

[0057] Key card 102, also known as a first UWB wireless transceiver, may include an omnidirectional antenna 121 and a corresponding processor for transmitting data using a technique that acts as a ranging initiator, which causes radio energy to spread across the UWB band of transmission channel 110. In some embodiments, key card 102 and / or vehicle-mounted UWB transceiver 104 include a transceiver circuitry and an associated antenna 141 for detecting and quantizing UWB signals.

[0058] The vehicle-mounted UWB transceiver 104, also known as a responder device or a second UWB wireless transceiver, acts as an antenna anchor or correlation device. It detects UWB pulses or correlation signals emitted by the key card 102 and forwards them to the BCM 110 or other onboard vehicle computers, position sensors, etc., to calculate the position of the key card 102. The signals exchanged with the key card 102 and output through the transmission channel 110 are part of the ranging operation, providing precise spatial awareness and relative positioning between the UWB devices 102 and 104 in system 10.

[0059] To detect and demodulate and / or decode signals received from key card 102, UWB transceiver 104 further includes antenna 141, which detects and quantizes UWB signals via transmission channel 110 and generates signals for locating key card 102. In some embodiments, antenna 141 may be similar to or the same as antenna 121 of key card 102 for performing UWB signal exchange. Antenna 141 may communicate with signal detector 142, which may be part of a radio frequency (RF) subsystem or an associated analog front end. Signal detector 142 in the analog domain may include analog-to-digital converters (ADCs) and digital-to-analog converters (DACs) for electronic communication with ranging processor 144, etc. For example, signal detector 142 detects and quantizes signals received by UWB transceiver 104 via antenna 141, such as energy levels, power spectral density, etc. More specifically, ranging processor 144 is preferably configured and arranged to detect and amplify UWB signals and generate at least one CIR based on the UWB signals. In doing so, the ranging processor 144 may be part of a baseband subsystem or the like, used to execute digital receiver algorithms to perform ToA measurements or related distance determinations or ranging sequence calculations.

[0060] In some embodiments, the UWB transceiver 104 includes a CIR feature processor 146 that extracts features from the CIR for analysis and subsequent processing of the filtered CIR using a machine learning classification algorithm, executed, for example, by a machine learning computer 148, to distinguish reflections in the CIR signal required to identify a first path in a normally dense multipath environment. In some embodiments, the machine learning computer 148 includes a dedicated hardware processor within a housing or enclosure 143 of the UWB transceiver 104 for executing some or all of the machine learning algorithms. The machine learning code is stored in the memory of the machine learning computer 148 of the transceiver 104 and executed by the hardware computer processor of the machine learning computer 148. In other embodiments, the CIR feature processor 146 is part of a remote dedicated computer that communicates electronically with the UWB transceiver 104 to receive and process CIR data. In other embodiments, the machine learning computer 148 may be located remotely from the UWB transceiver 104 and communicate with the CIR feature processor 146. Here, some parts of the machine learning algorithm can be executed in the CIR feature processor 146, while other parts of the machine learning algorithm are executed at a remote machine learning computer. In embodiments where the machine learning computer 148 is housed at or otherwise part of the UWB transceiver 104, a single machine learning classifier is executed for each transceiver 104 used as a UWB anchor to classify the CIR features of interest at the node. In embodiments where a single classifier is used for multiple anchors, the machine learning computer 148 is decoupled from the anchors, such that the single classifier exchanges decision data, etc., with each anchor it is connected to.

[0061] Figure 5 This is a flowchart illustrating a method 200 for locating a key card inside a vehicle cabin according to some embodiments. In describing method 200, reference is made to… Figure 3 and 4 Some or all of the components of the UWB communication system 10. Therefore, some or all of the steps of method 200 can be performed by... Figure 3 and 4 The components of the UWB communication system 10 are executed.

[0062] In frame 202, key card 102 is located within the cabin, cockpit, galley, or other enclosed or partially enclosed area of ​​a car, airplane, ship, or other vehicle having reflective surfaces that could cause interference or other signal distortion at a UWB transceiver 104 located within the cabin or associated enclosure. Key card 102 may be positioned in a specific area, such as the seats, floor, dashboard, etc., of the cabin or associated enclosure.

[0063] In block 204, a ranging operation is performed. In some embodiments, the initiation process is performed by key card 102, transceiver 104, and / or a third-party or intervention device before or during the ranging operation. For the purposes of describing method 200, reference is made to key card 102, although method 200 can be equally applied to transceiver 104. For example, reference is made to... Figure 3 The startup routine can be stored in the memory of the key card 102, but is not limited thereto, enabling the processor of the key card 102 to execute the stored routine. The routine may include a series of instructions, such as clearing the cache of the device 102 to bring the key card 102 into an operational state. The key card 102 is pre-configured, for example, by pre-programming via software, to achieve full device performance in a manner that produces the best possible performance for the UWB device. For example, the key card 102 can be configured with parameters, for example, conforming to the IEEE 802.15.4 standard, to allow communication between the initiator device 102 and the transceiver 104. When performing ranging operations, distance measurements, such as ToA, Time of Flight (ToF), Time Difference of Arrival (TDoA), Angle of Arrival (AoA), etc., are determined between the key card 102 and the transceiver 104 for applications such as positioning and location tracking.

[0064] In box 206, estimate one or more CIRs. For example, refer again... Figures 1-3 The CIR can be measured at one or more locations of the transceiver 104 at the key card 102 relative to the vehicle 12. The estimated CIR includes location-related features or characteristics that can be extracted from the CIR by the system 10 in box 208, such as the CIR feature processor 146 of the UWB transceiver 104 for subsequent processing.

[0065] In box 210, a machine learning algorithm is provided to analyze the distance measured according to the CIR and to combine the analyzed distance with features extracted from the CIR. Specifically, machine learning helps to locate a key card inside the vehicle cabin. In doing so, a supervised learning algorithm can be implemented in hardware and / or software, which identifies the location of the key card 102 during training. At least one CIR is processed to generate data for a classification algorithm. The CIR produces characteristics of a “channel”, and the CIR can be analyzed to determine the number of reflections, how close they are (in time and space), the intensity of the reflections, and how they combine at the receiver 104. Since the signal path to be received from the transmitter can vary greatly from different locations of the transmitter and receiver, the CIR “fingerprint” can differ from many locations. The machine learning algorithm will be able to extract features from the CIR and use supervised learning to “learn” how the CIR varies between locations and use the CIR characteristics to perform localization. Unlike deep learning algorithms, the CIR feature extraction machine learning algorithm does not consist of a large number of algorithmic layers, each of which provides a different interpretation of the data it processes, and the algorithmic layers are arranged in a complex, processor-intensive network. Therefore, the smaller processing requirements required to perform CIR feature extraction machine learning algorithms are necessary compared to the larger processing requirements required for multi-layer deep learning algorithms.

[0066] In some embodiments, box 210 includes identifying a classification algorithm that is applicable and practically executable. The selected classification algorithm receives CIR features as input and generates an estimated location of the keycard as output at box 212. Specifically, the keycard is located in the in-vehicle environment in response to a combination of CIR feature extraction and machine learning processing. Example classification algorithms may include, but are not limited to, neural networks, support vector machines (SVMs), linear discriminant methods, decision trees, k-nearest neighbor (KNN), etc. For example, KNN can be applied to a dataset of CIR features, estimated distances, and various location-related measurements, and the location of the keycard inside the vehicle can be estimated in a new dataset with CIR measurements input to this classification algorithm, which in turn outputs a location estimate.

[0067] Figure 6 This is a flowchart illustrating a method 300 for training a machine learning classification algorithm to locate a key card, according to some embodiments. In describing the machine learning technique 300, reference is made to… Figure 3 and 4 Some or all of the components of the UWB communication system 10. Therefore, some or all of the steps of method 300 can be performed by... Figure 3 and 4 The components of the UWB communication system 10 are executed.

[0068] Boxes 302-306 of Method 300 may be part of a preprocessing function, where the data is arranged in order of the training and test datasets for the machine learning algorithm. Here, CIR features are extracted, normalized, filtered, and labeled, for example, by the CIR feature processor 146 of UWB anchor 104. Boxes 308-314 may be part of a learning function. Other machine learning steps, such as model evaluation and prediction, may follow Method 300.

[0069] Specifically, in box 302, at least one CIR is obtained. For example, this can be done separately for... Figure 3 and 4 The data exchange between the key card 102 and the UWB anchor 104 shown calculates the CIR. For example, as part of a ranging sequence, a message including a UWB preamble is output from a ranging initiator device, such as the key card 102, and the message is received by a responder device, such as the UWB anchor 104, which processes the preamble to generate the CIR.

[0070] In box 304, the CIR can be filtered to reduce the complexity associated with subsequent extraction step 306. For example, it can be filtered... Figure 1 The relevant area 1 or Figure 2 The CIR portion outside the correlation region 2 is filtered to extract the contents of correlation regions 1 and 2, such as peaks, leading edges, and energy levels.

[0071] In box 306, features of the CIR are extracted, including identifying and analyzing the filtered CIR and performing mathematical operations in the time or frequency domain. The identified and extracted features may include, but are not limited to, the number of peaks in the CIR, the first path peak / maximum peak ratio, the first path peak, the maximum peak distance, width and / or prominence, the CIR energy, the CIR spectral power, the first path peak skewness and kurtosis. Figure 7 A graph of the features is shown, including the degree of bulge (a), width (b), and number of peaks (c) of the first path extracted in the preprocessing step. In some embodiments, the features may include estimated distances. Regardless of the accuracy of the estimated distance, machine learning classification algorithms can be trained with the estimated distance, etc. In some embodiments, CIR features are labeled to locate the keycard in a manner sufficient to enable the machine learning algorithm to learn and process the recognized labels, and to classify other CIR data based on each feature it learns from the labels.

[0072] Machine learning algorithms applied to extracted CIR features do not require structured or labeled data to classify the CIRs. For example, embodiments include generating structured data so that the machine learning algorithm learns, for instance, labeled data used for training. As described herein, a classifier can classify keycard data based on features it has learned through labels. This differs from applications that rely on process-intensive deep learning algorithms to output their data inputs through different layers of a hierarchical network.

[0073] In decision diamond 308, it is determined whether the classification algorithm has been trained. In some embodiments, supervised learning is performed, wherein the algorithm identifies keycard locations during training and, as previously described in boxes 304 and 306, processes the CIR to generate data for the classification algorithm. Here, supervised learning requires that the possible outputs of the algorithm are known, and that the CIR data used to train the algorithm has been labeled.

[0074] In some embodiments, a model must be generated for the classification algorithm, and thus a training step is performed (box 310), followed by a model generation process (box 312). Once the model is generated, the classifier can output labels as part of the new data. In other embodiments, the training algorithm receives features of the labels as input at specified locations (from box 306) and returns a model for a selected classifier as output, which is used to classify the data (box 314). Example classifiers applied to the extracted CIR features may include, but are not limited to, support vector machines, k-nearest neighbors (KNN), Naive Bayes, decision trees, random forests, neural networks, etc. In some embodiments, after the model training / learning algorithm process, the model is evaluated against a test dataset provided as part of the above preprocessing steps and directly input into the model as part of the model evaluation process.

[0075] Another feature of the learning step in method 300 according to an embodiment of the present invention is that fewer classifiers are selected for initial testing, and therefore fewer UWB anchors 104 are required. In some embodiments, an anchor can collect data from other anchors and then locally execute a machine learning algorithm at a representative anchor. The representative anchor can then output request signals, etc., to the BCM 106 or other automotive control processor to unlock doors, etc. Compared to the complex decision-making operations of deep learning algorithms, decision trees or other machine language programs can be implemented with lower complexity and higher simplicity, and the complex decision-making operations can therefore be processed at specific UWB anchors 104. Therefore, decisions can be made by a single anchor classifier at each individual anchor.

[0076] As should be understood, the disclosed embodiments include at least the following. In one embodiment, a system for locating an ultra-wideband (UWB) device includes: a UWB transceiver that identifies and extracts features of at least one channel impulse response (CIR); and a dedicated processor that applies a machine learning classification process to the extracted CIR features to locate the UWB device in a vehicle.

[0077] Alternative embodiments of the system include one or any combination of the following features: The UWB device is a key card. The UWB transceiver includes a CIR feature processor that extracts features of the CIR for analysis and subsequent processing via the machine learning classification process. A machine learning computer distinguishes reflections in the CIR required to identify a first path in a dense multipath environment of the vehicle. The CIR feature processor further normalizes, filters, and / or labels the CIR during the classification phase of the process performed by the system. The UWB transceiver includes: an antenna; a ranging processor that generates the CIR based on UWB signals received from the UWB device by the antenna; and a signal detector that detects and quantizes the UWB signals received by the UWB transceiver. The UWB transceiver includes a housing, and the dedicated processor is housed within the housing. The dedicated processor applies a single classifier of the machine learning classification process to the extracted CIR features. The dedicated processor is configured and arranged to receive and process distance data about the UWB device for training the machine learning classification process to distinguish the location of the UWB device inside the vehicle.

[0078] In another embodiment, an ultra-wideband (UWB) wireless communication system includes: a key card that outputs a UWB signal; at least one anchor that participates in a ranging sequence together with the key card, the ranging sequence including generating at least one channel impulse response (CIR) based on the UWB signal and extracting features of the at least one CIR; and a dedicated processor that applies a machine learning classification process to the extracted CIR features to locate the key card.

[0079] Alternative embodiments of the system include one or any combination of the following features. The at least one anchor includes a plurality of anchor devices, each storing and processing a classifier for the machine learning classification process regarding the CIR features. The machine learning classification process selects the CIR features to train the classifier. The machine learning classification process executes a training algorithm and returns a model for the classifier as output, the classifier being used to classify data based on the CIR features. A machine learning computer distinguishes reflections in the at least one CIR required to identify a first path in a dense multipath environment of the vehicle. The at least one anchor device includes: an antenna; a ranging processor that generates the at least one CIR based on a UWB signal received by the antenna; a signal detector that detects and quantizes the UWB signal received by the UWB transceiver; and a CIR feature processor that extracts the features of the at least one CIR. The at least one anchor device includes a housing, and the dedicated processor is housed within the housing.

[0080] In another embodiment, a method for locating an ultra-wideband (UWB) device includes: outputting a UWB signal from the UWB device to a UWB transceiver; extracting features of at least one channel impulse response (CIR) of the UWB signal from the UWB transceiver; and applying a machine learning classification process to the extracted CIR features by a dedicated processor to locate the UWB device in a vehicle.

[0081] Alternative embodiments of the system include one or any combination of the following features. The method further includes: processing the CIR to distinguish reflections in the CIR of the UWB signal; and identifying a first path in a dense multipath environment of the UWB device. Processing the CIR further includes generating data for the machine learning classification process. The machine learning classification process executes a training algorithm to return a model of a classifier for classifying the CIR features. The machine learning classification process estimates the position of the UWB device within the vehicle cabin.

[0082] While the invention has been described herein with reference to specific embodiments, various modifications and changes may be made without departing from the scope of the invention as set forth in the appended claims. Therefore, the specification and drawings should be considered illustrative rather than restrictive, and all such modifications are contemplated to be included within the scope of the invention. It is not intended that any advantage, benefit, or solution to the problem described herein with reference to specific embodiments be construed as a critical, necessary, or essential feature or element of any or all claims.

[0083] Unless otherwise stated, terms such as “first” and “second” are used to distinguish, arbitrarily, the elements described by such terms. Therefore, these terms are not necessarily intended to indicate temporal or other priorities of such elements.

Claims

1. A system for locating ultra-wideband (UWB) devices, characterized in that, include: A UWB transceiver, the UWB transceiver including a channel impulse response (CIR) processor, the channel impulse response (CIR) processor being configured to identify and extract features of at least one channel impulse response (CIR), wherein the channel impulse response (CIR) processor is further configured to filter multiple features of at least one channel impulse response to provide a filtered channel impulse response (CIR). as well as A dedicated processor is configured to apply a machine learning classification process to a filtered CIR for analysis and subsequent processing of the machine learning classification process to locate the UWB device in a vehicle. The dedicated processor is configured to receive estimated distance data of the UWB device and train the machine learning classification process based on the estimated distance data and the filtered channel impulse response (CIR) to form a model indicating a specific location of the UWB device in the vehicle based on CIR inputs to the model.

2. The system according to claim 1, characterized in that, The UWB device is a key card.

3. The system according to claim 1, characterized in that, The machine learning computer distinguishes the reflections in the filtered CIR required to identify the first path in the dense multipath environment of the vehicle.

4. The system according to claim 1, characterized in that, The CIR feature processor further normalizes, filters, and / or labels the plurality of features of the at least one CIR.

5. The system according to claim 1, characterized in that, The UWB transceiver includes: antenna; A ranging processor, which generates the at least one CIR based on a UWB signal received from the UWB device by the antenna; and A signal detector that detects and quantizes the UWB signal received by the UWB transceiver.

6. The system according to claim 1, characterized in that, The UWB transceiver includes a housing, and the dedicated processor is housed within the housing.

7. The system according to claim 6, characterized in that, The dedicated processor applies a single classifier from the machine learning classification process to the filtered CIR features.

8. An ultra-wideband (UWB) wireless communication system, characterized in that, include: Key card, the key card outputs a UWB signal; At least one anchor device, the at least one anchor device participating together with the key card in a ranging sequence, the ranging sequence including generating at least one channel impulse response (CIR) based on the UWB signal, and the anchor device including a channel impulse response CIR processor configured to extract multiple features of the at least one CIR, wherein the channel impulse response CIR processor is further configured to filter the multiple features of the at least one channel impulse response to provide a filtered channel impulse response CIR; as well as A dedicated processor applies a machine learning classification process to a filtered CIR for analysis and subsequent processing of the machine learning classification process to locate the key card; wherein the dedicated processor is configured to receive estimated distance data from a UWB device and train the machine learning classification process based on the estimated distance data and the filtered channel impulse response (CIR) to form a model, the model indicating a specific location of the UWB device in the vehicle based on the CIR inputs input to the model.

9. A method for locating ultra-wideband (UWB) devices, characterized in that, include: The UWB device outputs a UWB signal to the UWB transceiver. Multiple features of at least one channel impulse response (CIR) of the UWB signal are extracted by the UWB transceiver; Filter at least one channel impulse response (CIR) multiple features to provide a filtered channel impulse response (CIR); as well as A dedicated processor applies a machine learning classification process to the filtered CIR features to locate the UWB device in the vehicle; The dedicated processor is configured to receive estimated distance data from the UWB device and train the machine learning classification process to form a model based on the estimated distance data and the filtered channel impulse response (CIR), the model indicating the specific location of the UWB device in the vehicle based on the CIR inputs input to the model.

Images