Wireless channel rendering using neural networks

Abstract

Certain aspects of the present disclosure provide techniques and apparatus for training and using a machine learning model to estimate a representation of a channel between a transmitter and a receiver in a spatial environment. An example method generally includes estimating a representation of a channel using a machine learning model trained to generate an estimated representation of the channel based on a position of a transmitter in a spatial environment, a position of a receiver in the spatial environment, and information about the spatial environment. One or more actions are taken based on the estimated representation of the channel.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. patent application serial number 17 / 935,006, filed on September 23, 2022, which is hereby incorporated by reference into this application. Background Art

[0003] Aspects of the present disclosure relate to using machine learning to estimate wireless channels using neural networks.

[0004] In wireless communication systems, information about wireless channels can be used for various purposes. For example, information about wireless channels can be used to identify various parameters used for communication between a transmitter and a receiver, such as beamforming parameters, directional beam selection, etc. Information about wireless channels can also be used to determine the layout of the spatial environment in which the transmitter and receiver are located, which in turn can be used for various purposes, such as detecting devices entering and exiting different areas (e.g., defined based on a radius from a specified device). Layout information and position estimates can also be used for many other purposes, such as emergency management within a spatial region, spatial optimization, etc.

[0005] The state of a wireless channel generally depends on various factors in the spatial environment. For example, a wireless channel may be affected by sources of radio frequency interference, such as interfering network entities. In another example, hard surfaces such as walls, support columns, etc., and materials in these environments may introduce attenuation and reflection of radio waves in radio frequency measurements obtained in the spatial region. Because the state of a wireless channel may depend on many factors that vary across different environments, it may be difficult to estimate the state of a wireless channel.

[0006] Therefore, what is needed is an improved technique for estimating wireless channels in a spatial environment. Summary of the invention

[0007] An aspect provides a method for estimating a wireless channel using a machine learning model. The example method generally includes estimating a representation of the channel using a machine learning model, the machine learning model being trained to generate an estimated representation of the channel based on a location of a transmitter in a spatial environment, a location of a receiver in the spatial environment, and information about the spatial environment. One or more actions are taken based on the estimated representation of the channel.

[0008] Another aspect provides a method for training a machine learning model to estimate a wireless channel. The example method generally includes generating a training data set including a plurality of samples. Typically, each sample corresponds to a location of a receiver in a spatial environment and a location of a transmitter in the spatial environment and includes a plurality of vectors, each respective vector representing a respective signal component received at the receiver while traversing the spatial environment. The machine learning model is trained to generate an estimated representation of the channel based on information about the spatial environment and the training data set.

[0009] Other aspects provide: a processing system configured to perform the aforementioned methods and those described herein; a non-transitory computer-readable medium comprising instructions that, when executed by one or more processors of the processing system, cause the processing system to perform the aforementioned methods and those described herein; a computer program product embodied on a computer-readable storage medium comprising code for performing the aforementioned methods and those further described herein; and a processing system comprising components for performing the aforementioned methods and those further described herein.

[0010] The following description and the associated drawings set forth in detail certain illustrative features of one or more implementations. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The drawings depict certain aspects of the one or more embodiments and therefore are not to be considered limiting of the scope of the present disclosure.

[0012] Figure 1 Depicted are signals transmitted between a transmitter and a receiver in a space environment.

[0013] Figure 2 An example pipeline for estimating a wireless channel using a machine learning model according to aspects of the present invention is depicted.

[0014] Figure 3 Depicted are example machine learning models for generating an estimated representation of a channel based on a transmitter's location in a spatial environment, a receiver's location in the spatial environment, and information about the spatial environment in accordance with aspects of the present disclosure.

[0015] Figure 4 Depicted are examples of minimizing or at least reducing the loss between an estimated representation of a channel and a ground-truth representation of the channel in accordance with aspects of the present disclosure.

[0016] Figure 5Depicted are example operations for estimating a representation of a channel based on information about a spatial environment, a location of a receiver in the spatial environment, and a location of a transmitter in the spatial environment, in accordance with aspects of the present disclosure.

[0017] Figure 6 Depicted are example operations for training a machine learning model to estimate a representation of a channel based on information about a spatial environment, a location of a receiver in the spatial environment, and a location of a transmitter in the spatial environment in accordance with aspects of the present disclosure.

[0018] Figure 7 Depicted are example implementations of a processing system on which a machine learning model is used to estimate a representation of a channel in a spatial environment in accordance with aspects of the present disclosure.

[0019] Figure 8 Depicted are example implementations of a processing system on which a machine learning model is trained to estimate a representation of a channel in accordance with aspects of the present disclosure.

[0020] To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements that are common to the figures. It is contemplated that elements and features of one embodiment may be beneficially incorporated in other embodiments without further recitation. DETAILED DESCRIPTION

[0021] Aspects of the present disclosure provide techniques and apparatus for estimating a channel between a transmitter and a receiver based on location information of the transmitter and the receiver and the layout of the spatial region in which the transmitter and the receiver operate.

[0022] Information about the channel, such as a channel estimate at a receiver, can be used for various tasks within a wireless communication system and / or within a system using devices communicating within a wireless communication system. For example, information about the channel can be used for signal management, such as beamforming or beam selection, for communications between a transmitter and a receiver. In another example, information about the channel can be used for various sensing tasks, such as position estimation, floor map estimation, etc. This information can in turn be used to determine how to use a spatial region, to generate virtual reality or extended reality scenes in a spatial region, for traffic management within a spatial region, for position estimation, etc. In yet another example, the learned positions of signal reflectors in a spatial environment and the signal reflection characteristics of these reflectors can be used to perform various tasks in a dynamic environment, such as autonomous driving or other autonomous operations in which the state of the environment changes continuously.

[0023] Given information about the location of a transmitter and the location of a receiver in a spatial environment, various techniques may be used to estimate information about the channel. For example, a neural statistical model implemented as a deep neural network may be used to generate an estimate of the channel based on various field measurements of the transmitter and the receiver. These models may allow the estimation of channels in various general scenarios with minimal computational overhead. However, the channel estimates generated by these models may reflect channel estimates in a general environment and may not take into account environment-specific characteristics that affect the channel state at the receiver. For example, these general models may not take into account reflections of signals from surfaces in the spatial environment, signal attenuation due to these reflections, signal attenuation and reflection characteristics of different materials used in the spatial environment, and the like.

[0024] Various aspects of the present disclosure provide techniques that allow for estimating channels based on spatially consistent machine learning models that take into account information about the environment in which the transmitter and receiver operate when estimating information about the channel. By doing so, various aspects of the present disclosure can generate accurate estimates of the channel between the transmitter and the receiver in any specified environment. These estimates of the channel may take into account different components of the wireless signal received at the receiver at different times due to different propagation paths in the spatial environment, and may ignore signal components that have minimal impact on the total received channel at the receiver. By accurately estimating the channel in the spatial environment, various aspects of the present disclosure can achieve improved accuracy in applications that use channel estimation to perform various tasks. For example, because the accurate channel estimates generated by the machine learning models discussed herein can be used to identify blockages within the spatial environment that degrade the wireless signal, these accurate channel estimates can allow for improved communications in various spatial environments and at various frequencies at which the transmitter and receiver communicate, to estimate the best beam used in high-frequency (e.g., millimeter wave (mmWave)) communications between the transmitter and the receiver, and the like. In addition, the machine learning models discussed in this article can allow the propagation of signals between transmitters and receivers to be modeled, which can allow information about the spatial environment, such as the presence of reflective or refractive surfaces in the spatial environment and the RF reflectivity and attenuation characteristics of these surfaces, to be obtained from the estimates of the channels generated by these machine learning models.

[0025] Example RF signal in a spatial region

[0026] Figure 1 Example signals transmitted between a transmitter and various receivers within a spatial environment are shown.

[0027] As shown, within a spatial environment 100, a transmitter 110 transmits signals to receivers 112, 114, and 116. Transmitter 110 may be, for example, a gNodeB in a cellular telecommunication system, and receivers 112, 114, and 116 may be user equipment (UE) served by the gNodeB.

[0028] Because the space environment 100 includes various surfaces from which signals can be reflected, such as walls, floors, ceilings, or other surfaces, signaling sent between the transmitter 110 and the receivers 112, 114, and 116 may include a line-of-sight (LOS) component and, in some cases, one or more non-LOS components. For example, as shown, the signal received at the receiver 112 includes a LOS component corresponding to path 1 102, a first non-LOS component corresponding to path 2 104, and a second non-LOS component corresponding to path 3 106, in which the receiver 112 receives reflections of the signal off a wall in the space environment 100, and in which the receiver 112 receives reflections of the signal off a floor in the space environment 100. Similarly, the signal received at the receiver 114 includes a LOS component and a non-LOS component, wherein the receiver 114 receives reflections of the signal off a wall in the space environment 100.

[0029] Because the receiver may receive signals that include both LOS components and non-LOS components, for any signal transmitted by transmitter 110, the total impulse response (or signal strength) f θ An impulse response for a LOS component and an impulse response for a non-LOS component may thus be included. As shown in impulse response curve 120 for receiver 112 and impulse response curve 130 for receiver 114, the LOS component may have the highest impulse response and the shortest delay between transmission by transmitter 110 and reception by an appropriate receiver, and the non-LOS components may have a lower impulse response and a longer delay between transmission and reception relative to the LOS components. Because signal strength decreases as distance increases, because different surfaces may introduce different amounts of signal attenuation into the non-LOS components, and the total distance traversed between the transmitter and the receiver, these non-LOS components may have a lower impulse response due to various factors, including signal attenuation from surfaces from which these signals are reflected.

[0030] In general, the global representation of a wireless channel can be modeled by the following equation:

[0031]

[0032] where y(t) is the global representation of the channel at time t, and a i represents the gain for the i-th path, and τi represents the delay for the i-th path. This global representation shows that the wireless channel is an aggregate of the signal received at the receiver for each path from the transmitter to the receiver (e.g., for both LOS and non-LOS components). However, this global representation may not allow modeling of the propagation characteristics of the signal on any specified path. More specifically, for a specified path, this global representation may not allow modeling of reflections and signal attenuation for any non-LOS component of the signal.

[0033] In order to account for various characteristics of a three-dimensional spatial environment that affect signal reflections and signal attenuation, and to allow channel estimation within various spatial environments based on signal measurements obtained for received signals within the spatial environment, various aspects of the present disclosure provide techniques for estimating a representation of a channel using a neural network based on the locations of a transmitter and a receiver within the spatial environment and a three-dimensional representation of the spatial environment. As discussed in further detail herein, the neural network can estimate the channel based on estimates of signal attenuation due to propagation distance and signal attenuation due to ray-surface interactions (e.g., signal absorption, reflection angle, etc.). The entire channel can be rendered based on these estimates so that the channel includes signal components that are believed to have been received by the receiver (e.g., points within a defined reception ellipsoid propagating to the receiver), and excludes signal components that have an insignificant effect on the entire channel due to attenuation below a threshold signal strength.

[0034] Example estimation of a wireless channel using a machine learning model

[0035] As discussed, within a wireless communication system, a signal may be represented as a series of components received at a receiver. Each component may, for example, correspond to a ray sent from a location within a transmitter's spatial environment that arrives at a receiving sphere, which is defined as the receiver's location within the spatial environment and a radius from that location. The characteristics of each signal component (such as received power, flight time, departure angle, and arrival angle, etc.) may be affected by various factors, such as the two-dimensional layout of the spatial environment (e.g., the location of walls or other reflective surfaces, etc.), the location of the transmitter in the spatial environment, and the location of the receiver in the spatial environment.

[0036] Typically, statistical channel models and generative channel models can estimate representations of channels with low computational overhead. However, these statistical channel models and generative channel models may not be portable. That is, statistical channel models and generative channel models may be trained or otherwise configured to estimate representations of channels within a specific spatial environment (e.g., a specific room or set of rooms within a building). Because these statistical channel models and generative channel models are tightly bound to a specific spatial environment, significant computational overhead may be incurred when estimating models of channels in various spatial environments.

[0037] Various aspects of the present disclosure provide techniques and apparatus that allow the use of machine learning models to estimate channels in various spatial environments. These machine learning models generally use information about the spatial environment in which the transmitter and receiver are located and the location of the transmitter and receiver in the spatial environment to render an estimate of the channel between the transmitter and the receiver. These machine learning models can do so without incurring the computational overhead of computationally complex operations to model the propagation of a large number of signal components in the spatial environment. Therefore, various aspects of the present disclosure provide efficient and portable techniques for estimating wireless channels across various spatial environments.

[0038] Figure 2 A pipeline 200 for estimating a wireless channel in a spatial environment using a machine learning model according to aspects of the present disclosure is shown. As shown, the pipeline 200 includes a renderer 210 and a minimization loss function 230. The renderer 210 generally represents a machine learning model that is trained to use information about the spatial environment (e.g., an image describing a two-dimensional layout of the spatial environment), a location x of a receiver (e.g., a user equipment (UE) in a wireless communication system), and a location x of a receiver (e.g., a user equipment (UE) in a wireless communication system). UE , and the location x of the transmitter (e.g., a base station such as a gNode B (gNB) in a wireless communication system) gNB The information of position x is used to generate an estimate of the channel 220. UE and x gNB can be represented as a vector and can be different from the following Figure 3 The scalar value of x included in the position vector in question.

[0039] The channel estimate 220 between the transmitter and the receiver may include a plurality of vectors, each vector corresponding to a signal component received at the receiver (e.g., when combined, representing one of a plurality of multipath components of the signal received at the receiver). Each signal component in the estimate 220 may be represented as a collection of scalar and vector data. For example, as shown, each signal component k may be represented as a power data vector a k (which may include the power of a signal component transmitted from a transmitter to a receiver and the power of the transmitted signal component received at the receiver), a scalar flight time component τ representing the elapsed time between transmission and reception of the signal component k , represents the departure angle component corresponding to the departure direction (DOD) of the signal component from the transmitter A vector of , and a component representing the angle of arrival (DOA) corresponding to the direction of arrival (DOA) of the signal component at the receiver In some aspects, the signal components in estimating 220 may also include m kInformation indicating a confidence level associated with a signal component. The signal component may be used, for example, to indicate the likelihood that the signal component is valid or otherwise has a meaningful impact on a rendered channel at a receiver.

[0040] The renderer 210 can be trained for various scenarios in various spatial environments based on the true data 240. The true data 240 generally corresponds to an actual representation of a channel between a transmitter (e.g., a gNB) and a receiver (UE) in a specific spatial environment, and also includes multiple signal components. Typically, the true data 240 is used to train the machine learning model so that during the training process, the estimate generated by the machine learning model can be compared to the true data 240. The difference between the estimate generated by the machine learning model and the true data 240 can be used to further train the machine learning model so that the machine learning model can learn to generate an estimate that is closer to the true data 240, and thus cause the machine learning model to be assumed to generate accurate estimates for various scenarios. As with the estimate Each signal component k comprises a power data vector a k , scalar flight time component τ k , a vector representing the departure angle component and the vector representing the angle of arrival component That is, the estimation can be expressed by the following equation To train the renderer 210, the minimization loss function 230 may be configured to compute the loss between the estimate 220 of the channel in the spatial environment and the actual representation of the channel defined in the ground truth data 240. And the loss is used when performing subsequent training rounds for the machine learning model used by the renderer 210, as discussed in further detail below. Minimize loss function 230 can minimize or at least reduce the loss between the estimate 220 and the true value data h240. When minimizing or at least reducing the loss, minimize loss function 230 can identify a global minimum, a local minimum, or at least a reduction in loss relative to the loss calculated for the estimate generated by the machine learning model in the previous round of training. For example, the loss function can be expressed as The minimization loss function 230 may be represented by the following equation:

[0041] In some aspects, the information about the spatial environment input into the renderer 210 can be an image depicting the layout of the spatial environment. The image can be, for example, a two-dimensional image showing the layout of rooms or other discrete spaces in the spatial environment, openings between rooms or other discrete spaces in the spatial environment, and the boundaries of the spatial environment. In order to use this information as an input to a machine learning model, the image depicting the layout of the spatial environment can be converted into a representation that can be used in the machine learning model, such as an embedding of an image generated by a model (or sub-model) trained to embed image data into a different data space.

[0042] Figure 3 An example renderer 210 is shown in which a machine learning model is used to generate an estimated representation of a channel based on a transmitter's location in a spatial environment, a receiver's location in the spatial environment, and information about the spatial environment in accordance with aspects of the present disclosure.

[0043] As shown, the renderer 210 includes a first machine learning model 310 and a second machine learning model 320. φ The first machine learning model 310 is generally a machine learning model trained to generate an embedded representation of information about the environment in which the transmitter and receiver communicate and for which the channel is to be estimated. In some aspects, the machine learning model can be the encoder portion of a variational autoencoder (VAE) that is trained to generate a vector representing the spatial environment in which the signal component propagates from the transmitter to the receiver. The vector can be input into the second machine learning model 320 to adjust the rendering of the channel based on the layout of the environment in which the transmitter and receiver operate.

[0044] In some aspects, the first machine learning model 310 can be trained on a training dataset of multiple two-dimensional floor map images representing layouts of different spatial environments. In aspects where the first machine learning model 310 is a variational autoencoder, the first machine learning model 310 can be trained as an encoder e φ and decoder d φ The encoder φ and decoder d φ For example, it can be a multilayer perceptron (MLP) implemented as two layers of rectified linear units. In general, the encoder e can be trained φ To input x i (e.g., a 2D representation of a spatial environment) is encoded into an embedded representation Where D represents the embedding representation z i The dimension of the space in which it is located. The embedding represents z i It can be a vector different from the scalar z (e.g., the position on the z axis) in the position vector described below. In general, D can be set so that the embedding representation zi is located in a high-dimensional space to improve the accuracy of the rendered channels and allow the rendered channels to be rendered along dimensions larger than the number of dimensions in the physical space. The decoder d φ It is usually trained to extract the φ The input x i The embedded representation z encoded into it i Generate input x i Reorganization The variational autoencoder can be trained to minimize or at least reduce i and The loss between.

[0045] The second machine learning model 320 generally uses an embedding representation z i and the receiver position information (designated as P UE = (x, y, z, t, ...)) and the transmitter position information (designated as P gNB =(x, y, z, t, ...)) to render a continuous representation of the channel between the transmitter and the receiver, or otherwise generate an estimate of the channel (e.g., including a channel impulse response (CIR), identification of multipath components in the channel, etc.). In some aspects, the second machine learning model 320 can be a coordinate MLP configured to generate a signal output from a set of spatial coordinates. The MLP can include multiple layers and use rectifier linear unit activations.

[0046] In some aspects, the coordinate information associated with the location of the receiver and the location of the transmitter can be converted into a higher dimensional representation by transforming γ. In some aspects, the transform can be a Fourier transform that allows the generation of higher dimensional data from a data set of lower dimensional data. In some aspects, the Fourier transform can be an identity mapping such that γ(x)=x. In another example, the Fourier transform can be represented by the following equation:

[0047]

[0048] in represents a random Gaussian matrix that maps a low-dimensional coordinate x to a high-dimensional vector γ(x).

[0049] In some aspects, the output of the renderer 210 may include various components, such as a channel h defined according to time τ and all times t of the rendered signal, a channel impulse response, multipath components, and / or other data that may provide information about the channel between the transmitter and the receiver.

[0050] In some aspects, supervised learning techniques can be used to train the second machine learning model 320. The data set used to train the machine learning model may include data generated via a ray tracing software package or real-life data obtained along a path through the space environment. For example, the data set used to train the machine learning model may include data obtained for multiple discrete points within the space environment corresponding to grid positions in the space environment. Assume, for example, a space environment that is 60 meters wide by 30 meters deep. The data set used to train the machine learning model may include data obtained for blocks of size per square meter in the space environment, resulting in a data set of 1,800 samples of the space environment and a fixed transmitter position in the space environment. This process can be repeated for multiple different transmitter positions in the space environment to allow the second machine learning model 320 to be trained using a richer, more detailed data set.

[0051] Figure 4 An example 400 of minimizing or at least reducing the loss between an estimated representation of a channel and a true representation of the channel in accordance with aspects of the present disclosure is shown.

[0052] As discussed, the specified The estimated channel component of is denoted and designated as s l The true value representation of the channel component may include multiple data components, where some of the data components are scalar components and some of the data components are vector components. Corresponding to a set represented as including multiple paths Each path s l includes various information defining or otherwise providing information about a signal received at a receiver, including information corresponding to s l The received power, flight time, departure angle information and arrival angle information of the signal. Corresponding to the estimated channel, it is also represented as a set of multiple paths

[0053] Typically, the difference between the scalar components can be computed as an estimated representation of the channel The scalar component in the channel is represented by the estimated representation s l The mean square error (MSE) between the corresponding scalar components in . At the same time, the difference between the vector components can be calculated as the estimated representation of the channel The vector components in and the true value representation of the channel s l The cosine distance between the corresponding vector components in . The channel estimate is expressed as and the truth representation s of the channel l The total distance d460 between the two channels can be expressed as an estimate of and the truth representation of the channel sl The sum of each distance calculated between the corresponding components in .

[0054] For example, as shown in the figure, the estimated representation of the channel includes a received power scalar value a k (which may be the power of the signal component received at the receiver), a scalar time-of-flight component τ representing the elapsed time between transmission and reception of the signal component k , represents the departure angle component The vector of the angle of arrival In some aspects, the estimated representation of the channel may also include a two-dimensional Cartesian coordinate associated with the phase angle of the received signal component, which may be defined as It can be defined according to the following expression

[0055]

[0056] In some aspects, the departure angle component can be and the angle of arrival component Obtain three-dimensional Cartesian coordinates. For a specified departure angle θ or arrival angle and phase shift φ, the three-dimensional Cartesian coordinates can be defined according to the following expression:

[0057]

[0058] Therefore, as shown in the figure, the estimated channel components are represented by and the true value representation s of the channel components l The distance between may be the sum of the received power difference 410, the angular power difference 420, the flight time difference 430, the departure angle difference 440, and the arrival angle difference 450. The received power difference 410 and the flight time difference 430 may be calculated as mean squared errors because these components may be included in the estimated representation of the channel component. and the true value representation of the channel component s l At the same time, the angular power difference 420, the departure angle difference 440 and the arrival angle difference 450 can be calculated as cosine difference values, because these components can be represented in the estimated representation of the channel component and the true value representation of the channel component s l as vector components.

[0059] In some aspects, the estimated representation of the channel may be calculated based on a chamfer distance between the estimated representation of the channel and the true representation of the channel and a binary cross entropy (BCE) calculated between the confidence scores m of the channel components in the estimated representation of the channel and the true representation of the channel. The truth representation of the channel The loss function between . The chamfer distance can be expressed by the following equation:

[0060]

[0061] The total loss function can be expressed by the following equation:

[0062]

[0063] where m corresponds to a fixed-dimensional vector with the confidence scores for each path, and m gt corresponds to a truth vector having a value of 1 assigned to valid paths and a value of 0 assigned to invalid paths (eg, paths that do not result in reception of a signal component at the receiver).

[0064] In some aspects, in order to train a machine learning model and minimize (or at least reduce) the overall loss, the estimated representations of the channel components and the true value representations of the channel components can be organized in a matrix. For a pairing σ(m k ) is less than a threshold, the signal component pairing can be masked to reduce the amount of data processed in the training machine learning model. In some aspects, the pairing can also be padded with zero values ​​to result in a matrix of valid pairings between the true value and the estimated signal component that is appropriately sized for training the machine learning model.

[0065] Example channel rendering using a machine learning model

[0066] Figure 5 Example operations 500 are shown for estimating a representation of a channel based on information about a spatial environment, a location of a receiver in the spatial environment, and a location of a transmitter in the spatial environment in accordance with aspects of the present disclosure. The information may be, for example, provided by a user equipment in a wireless communication system on which a machine learning model is deployed, such as Figure 1 The UE 104 is shown to perform the operations 500 .

[0067] As shown, operation 500 begins at block 510, where a machine learning model is used to estimate a representation of a channel. The machine learning model is generally trained to generate an estimated representation of a channel based on the position of a transmitter in a spatial environment, the position of a receiver in the spatial environment, and information about the spatial environment. Information about the spatial environment may include, for example, a two-dimensional representation of the spatial environment (e.g., a floor map). The two-dimensional representation may include boundary information of the spatial environment, the position of a wall or other reflective / refractive surface in the spatial environment, and the like.

[0068] In some aspects, the machine learning model may include a first sub-model and a second sub-model. The first sub-model may be configured to encode information about the spatial environment into an embedded representation of the spatial environment. The second sub-model may be configured to generate an estimated representation of the channel as a continuous representation based on the embedded representation of the spatial environment, the position of the transmitter in the spatial environment, and the position of the receiver in the spatial environment. The first sub-model may be a variational autoencoder, and the second sub-model may be a multi-layer perceptron.

[0069] In some aspects, the machine learning model can be a model trained to minimize or at least reduce the chamfer distance between the estimated state of the channel and the actual state of the channel. The minimized (or reduced) chamfer distance between the estimated state of the channel and the actual state of the channel can be minimized (or at least reduced) on a per signal component basis, and the chamfer distance between the estimated state of the channel and the actual state of the channel can be calculated as the sum of the differences between corresponding components in the vectors representing the same signal component in the estimated state of the channel and the actual state of the channel.

[0070] In some aspects, the machine learning model can be a model trained to minimize or at least reduce a loss between an estimated state of a channel and an actual state of the channel, the loss comprising a sum of a chamfer distance and a weighted binary cross entropy between an estimated representation of the channel and the actual state of the channel.

[0071] In some aspects, a higher dimensional representation of the location of the transmitter in the spatial environment and a higher dimensional representation of the location of the receiver in the spatial environment are generated based on a Fourier transform or other transform. Typically, the higher dimensional representation can have more dimensions than two dimensions, in which the locations of the receiver and transmitter in the spatial environment are represented on a plane, or can have more dimensions than three dimensions, in which the locations of the receiver and transmitter in the spatial environment are plotted in a three-dimensional space. The higher dimensional representation can be used as an input into a second sub-model to generate an estimated representation of the channel.

[0072] In some aspects, the estimated representation of the channel includes a plurality of vectors. Each vector generally corresponds to a respective signal component estimated to be received at the receiver. The vectors include received power, flight time, departure angle information, and arrival angle information of the respective signal component associated with the respective vector.

[0073] At box 520, operation 500 continues to take one or more actions based on the estimated representation of the channel. In some aspects, taking one or more actions includes generating a graphical rendering of the estimated representation of the channel in a three-dimensional representation of the spatial environment. In some aspects, taking one or more actions includes selecting one or more beams for communication between a transmitter and a receiver based on the estimated representation of the channel. In some aspects, taking one or more actions includes detecting the presence and position of an object in the spatial environment based on the estimated representation of the channel. In some aspects, taking one or more actions includes identifying material properties of the spatial environment based on the estimated representation of the channel.

[0074] In general, aspects of the present disclosure can provide accurate and computationally efficient predictions of channels in wireless networks. These predictions can be spatially consistent and can accurately reflect channel states in a spatial environment. Because the machine learning models discussed herein can provide accurate and computationally efficient channel predictions, these predictions can be used to improve communications between transmitting devices and receiving devices (e.g., by detecting blockages in the spatial environment that may adversely affect signals sent between the transmitting device and the receiving device, predicting the best beam or set of beams for communication between devices, etc.). In addition, the machine learning models discussed herein can use sensed data from transmitters and receivers and other sources (e.g., environmental maps, etc.) to accurately and efficiently render estimates of channels between devices in a spatial environment.

[0075] Figure 6 Depicted are example operations 600 for training a machine learning model to estimate a representation of a channel based on information about a spatial environment, a location of a receiver in the spatial environment, and a location of a transmitter in the spatial environment, in accordance with aspects of the present disclosure.

[0076] As shown, operation 600 begins at block 610 by generating a training data set including a plurality of samples. Each sample may correspond to a location of a receiver in a spatial environment and a location of a transmitter in the spatial environment. Each sample generally includes a plurality of vectors representing signal components received at a receiver at a particular location in the spatial environment.

[0077] In some aspects, each vector of the plurality of vectors includes received power, time of flight, angle of departure information, and angle of arrival information for a respective signal component associated with the respective vector.

[0078] In some aspects, the location of the receiver and the location of the transmitter can be high-dimensional representations of lower-dimensional location information of the receiver and the transmitter. These higher-dimensional representations of the location of the transmitter in the spatial environment and the location of the receiver in the spatial environment can be generated based on Fourier transforms or other transforms. Typically, the higher-dimensional representation can have more dimensions than two dimensions, in which the location of the receiver and the transmitter in the spatial environment is represented on a plane, or can have more dimensions than three dimensions, in which the location of the receiver and the transmitter in the spatial environment is plotted in three-dimensional space.

[0079] At block 620, operation 600 continues with training the machine learning model to generate an estimated representation of the channel based on information about the spatial environment and the training data set. For other aspects, the operation may involve generating a machine learning model for estimating a representation of the channel. In this case, the generating may include training the machine learning model.

[0080] In some aspects, training the machine learning model includes training a first sub-model and a second sub-model. The first sub-model can be trained to encode information about the spatial environment into an embedded representation of the spatial environment. The second sub-model can be trained to generate an estimated representation of the channel as a continuous representation based on the embedded representation of the spatial environment, the position of the transmitter in the spatial environment, and the position of the receiver in the spatial environment. The first sub-model can be a variational autoencoder, and the second sub-model can be a multi-layer perceptron. In some aspects, the first sub-model can be trained using unsupervised learning techniques, and the second sub-model can be trained using supervised learning techniques.

[0081] In some aspects, the machine learning model can be a model trained to minimize or at least reduce the chamfer distance between the estimated state of the channel and the actual state of the channel. The minimized (or reduced) chamfer distance between the estimated state of the channel and the actual state of the channel can be minimized (or at least reduced) on a per signal component basis, and the chamfer distance between the estimated state of the channel and the actual state of the channel can be calculated as the sum of the differences between corresponding components in the vectors representing the same signal component in the estimated state of the channel and the actual state of the channel.

[0082] In some aspects, training the machine learning model can include masking components in the estimated representation of the channel for components having confidence scores below a threshold confidence score. Typically, the masked components can include components that have a low probability of meaningfully contributing to the channel received at the receiver. By masking these components, the machine learning models can be trained on a set of correlated signal components that can be significantly smaller than a set of total signal components that can be generated by a transmitter in a spatial environment, which can reduce the computational cost of training these machine learning models.

[0083] In some aspects, the machine learning model can be a model trained to minimize or at least reduce a loss between an estimated state of a channel and an actual state of the channel, the loss comprising a sum of a chamfer distance and a weighted binary cross entropy between an estimated representation of the channel and the actual state of the channel.

[0084] Example processing system for channel rendering using machine learning models

[0085] Figure 7 An example processing system 700 is depicted for estimating channels in a spatial environment based on a machine learning model, such as described herein, for example, with respect to Figure 5 described.

[0086] Processing system 700 includes a central processing unit (CPU) 702, which in some examples may be a multi-core CPU. Instructions executed at CPU 702 may be loaded, for example, from a program memory associated with CPU 702, or may be loaded from memory 724.

[0087] The processing system 700 also includes additional processing components customized for specific functions, such as a graphics processing unit (GPU) 704 , a digital signal processor (DSP) 706 , a neural processing unit (NPU) 708 , a multimedia processing unit 710 , and a wireless connectivity component 712 .

[0088] An NPU, such as 708, is generally a dedicated circuit configured to implement control and arithmetic logic for executing machine learning algorithms, such as algorithms for processing artificial neural networks (ANNs), deep neural networks (DNNs), random forests (RFs), etc. An NPU is sometimes alternatively referred to as a neural signal processor (NSP), a tensor processing unit (TPU), a neural network processor (NNP), an intelligence processing unit (IPU), a vision processing unit (VPU), or a graphics processing unit.

[0089] The NPU, such as 708, is configured to accelerate the execution of common machine learning tasks, such as image classification, machine translation, object detection, and various other estimation models. In some examples, multiple NPUs may be instantiated on a single chip, such as a system on a chip (SoC), while in other examples, the NPU may be part of a dedicated neural network accelerator.

[0090] The NPU can be optimized (or specialized) for training or inference, or in some cases configured to balance performance between the two. For NPUs that can perform both training and inference, the two tasks can generally still be performed independently.

[0091] NPUs designed to accelerate training are generally configured to accelerate the optimization of new models, which is a highly computationally intensive operation that involves inputting an existing data set (usually labeled or labeled), iterating on the data set, and then adjusting model parameters (such as weights and biases) to improve model performance. Typically, optimization based on error estimates involves passing back through the layers of the model and determining gradients to reduce the estimation error. The degree to which the model is optimized can be related to the number of rounds that the machine learning model is trained. Each additional round of training can also optimize the machine learning model to improve the accuracy of the model in generating inferences, which can be measured based on the difference between the estimate generated by the machine learning model and the true value data for the specified input into the machine learning model.

[0092] NPUs designed to accelerate inference are generally configured to operate on complete models. Thus, such NPUs can be configured to input new data slices and quickly process them through the trained model to generate model outputs (e.g., inferences).

[0093] In one specific implementation, the NPU 708 is part of one or more of the CPU 702 , the GPU 704 , and / or the DSP 706 .

[0094] In some examples, the wireless connectivity component 712 may include, for example, subcomponents for third generation (3G) connectivity, fourth generation (4G) connectivity (e.g., 4G LTE), fifth generation connectivity (e.g., 5G or NR), Wi-Fi connectivity, Bluetooth connectivity, and other wireless data transmission standards. The wireless connectivity component 712 is further connected to one or more antennas 714.

[0095] The processing system 700 may also include one or more sensor processing units 816 associated with any manner of sensors, one or more image signal processors (ISPs) 718 associated with any manner of image sensors, and / or a navigation processor 720, which may include satellite-based positioning system components (e.g., GPS or GLONASS) and inertial positioning system components.

[0096] The processing system 700 may also include one or more input and / or output devices 722, such as a screen, a touch-sensitive surface (including a touch-sensitive display), physical buttons, speakers, microphones, and the like.

[0097] In some examples, one or more of the processors of processing system 700 may be based on the ARM or RISC-V instruction set.

[0098] The processing system 700 also includes a memory 724, which represents one or more static and / or dynamic memories, such as dynamic random access memory, flash-based static memory, etc. In this example, the memory 724 includes computer-executable components that can be executed by one or more of the aforementioned processors of the processing system 700.

[0099] Specifically, in this example, memory 724 includes a channel representation estimation component 724A, a take action component 724B, and a machine learning model component 724C. The depicted components, as well as other components not depicted, may be configured to perform various aspects of the methods described herein.

[0100] Generally, the processing system 700 and / or its components may be configured to perform the methods described herein.

[0101] It is worth noting that in other aspects, aspects of the processing system 700 can be omitted, such as where the processing system 700 is a server computer, etc. For example, in other aspects, the multimedia processing unit 710, the wireless connectivity component 712, the sensor processing unit 716, the ISP 718, and / or the navigation processor 720 can be omitted. In addition, aspects of the processing system 700 can be distributed, such as training a model and using the model to generate inferences, such as user-verified estimates.

[0102] Figure 8 An example processing system 800 for training a machine learning model to estimate channels in a spatial environment, such as described herein, for example, Figure 6 described.

[0103] The processing system 800 includes a central processing unit (CPU) 802, which in some examples may be a multi-core CPU. The processing system 800 also includes additional processing components customized for specific functions, such as a graphics processing unit (GPU) 804, a digital signal processor (DSP) 806, and a neural processing unit (NPU) 808. The CPU 802, GPU 804, DSP 806, NPU 808, and input and / or output devices 822 may be similar to those described above with respect to Figure 8 A CPU 802 , a GPU 804 , a DSP 806 , an NPU 808 , and input and / or output devices 822 are discussed.

[0104] In some examples, the wireless connectivity component 812 may include, for example, subcomponents for third generation (3G) connectivity, fourth generation (4G) connectivity (e.g., 4G LTE), fifth generation connectivity (e.g., 5G or NR), Wi-Fi connectivity, Bluetooth connectivity, and other wireless data transmission standards. The wireless connectivity component 812 may be further connected to one or more antennas (not shown).

[0105] In some examples, one or more of the processors of processing system 800 may be based on the ARM or RISC-V instruction set.

[0106] The processing system 800 also includes a memory 824, which represents one or more static and / or dynamic memories, such as dynamic random access memory, flash-based static memory, etc. In this example, the memory 824 includes computer-executable components that can be executed by one or more of the aforementioned processors of the processing system 800.

[0107] Specifically, in this example, the memory 824 includes a data set generation component 824A and a model training component 824B. The depicted components, as well as other components not depicted, may be configured to perform various aspects of the methods described herein.

[0108] Generally, the processing system 800 and / or its components may be configured to perform the methods described herein.

[0109] Sample Clauses

[0110] Specific implementation details for each aspect are described in the following numbered clauses.

[0111] Clause 1: A computer-implemented method comprising: estimating a representation of a channel using a machine learning model, wherein the machine learning model is trained to generate an estimated representation of the channel based on a location of a transmitter in a spatial environment, a location of a receiver in the spatial environment, and information about the spatial environment; and taking one or more actions based on the estimated representation of the channel.

[0112] Clause 2: The method of clause 1, wherein the information about the spatial environment comprises a two-dimensional representation of a layout of the spatial environment.

[0113] Clause 3: A method according to clause 1 or 2, wherein the machine learning model comprises: a first sub-model, the first sub-model being configured to encode the information about the spatial environment as an embedded representation of the spatial environment; and a second sub-model, the second sub-model being trained to generate the estimated representation of the channel as a continuous representation based on the embedded representation of the spatial environment, the position of the transmitter in the spatial environment, and the position of the receiver in the spatial environment.

[0114] Clause 4: The method according to Clause 3 further includes: generating a higher dimensional representation of the position of the transmitter in the spatial environment based on Fourier transform; generating a higher dimensional representation of the position of the receiver in the spatial environment based on Fourier transform; and inputting the higher dimensional representation of the transmitter in the spatial environment and the higher dimensional representation of the position of the receiver in the spatial environment into the second sub-model.

[0115] Clause 5: A method according to clause 3 or 4, wherein the first sub-model comprises a variational autoencoder.

[0116] Clause 6: A method according to any one of clauses 1 to 5, wherein the estimated representation of the channel comprises a plurality of vectors, each respective vector of the plurality of vectors corresponding to a respective signal component estimated to be received at the receiver, each respective vector comprising information identifying the received power, flight time, departure angle and arrival angle of the respective signal component associated with the respective vector.

[0117] Clause 7: A method according to any one of clauses 1 to 6, wherein the machine learning model comprises a model trained to minimize the chamfer distance over time between an estimated state of the channel and an actual state of the channel.

[0118] Clause 8: A method according to any one of clauses 1 to 7, wherein the machine learning model comprises a model trained to minimize the loss between an estimated state of the channel and an actual state of the channel, the loss comprising the sum of a chamfer distance between the estimated state of the channel and the actual state of the channel and a weighted binary cross entropy.

[0119] Clause 9: The method of any of clauses 1 to 8, wherein taking one or more actions comprises generating a graphical rendering of the estimated representation of the channel in a representation of the spatial environment.

[0120] Clause 10: The method of any of clauses 1 to 9, wherein taking one or more actions comprises selecting one or more beams for communicating between the transmitter and the receiver based on the estimated representation of the channel.

[0121] Clause 11: The method of any of clauses 1 to 10, wherein taking one or more actions comprises detecting the presence and location of an object in the spatial environment based on the estimated representation of the channel.

[0122] Clause 12: The method of any of clauses 1 to 11, wherein taking one or more actions comprises identifying material properties of the spatial environment based on the estimated representation of the channel.

[0123] Clause 13: A computer-implemented method comprising: generating a training data set comprising a plurality of samples, wherein each sample corresponds to a position of a receiver in a spatial environment and a position of a transmitter in the spatial environment, and each sample comprises a plurality of vectors, each respective vector representing a respective signal component received at the receiver at a particular position in the spatial environment; and training a machine learning model to generate an estimated representation of a channel based on information about the spatial environment and the training data set.

[0124] Clause 14: A method according to clause 13, wherein training the machine learning model comprises: training a first sub-model to generate an embedded representation of the spatial environment based on the information about the spatial environment; and training a second sub-model based on the embedded representation of the spatial environment and the training data set to generate the estimated representation of the channel.

[0125] Clause 15: The method of clause 14, wherein the first sub-model comprises a variational autoencoder.

[0126] Clause 16: A method according to any of clauses 14 or 15, wherein the first sub-model is trained using an unsupervised learning technique and the second sub-model is trained using a supervised learning technique.

[0127] Clause 17: A method as described in any of clauses 13 to 16, wherein each respective vector comprises information identifying a received power, a time of flight, a departure angle and an arrival angle of the respective signal component associated with the respective vector.

[0128] Clause 18: A method according to any one of clauses 13 to 17, wherein: the position of the transmitter in the spatial environment includes a high-dimensional representation of the position of the transmitter generated based on a Fourier transform applied to the three-dimensional position of the transmitter, and the position of the receiver in the spatial environment includes a high-dimensional representation of the position of the receiver generated based on a Fourier transform applied to the three-dimensional position of the receiver.

[0129] Clause 19: A method according to any one of clauses 13 to 18, wherein training the machine learning model comprises training the machine learning model to minimize the loss between the estimated state of the channel and the actual state of the channel.

[0130] Clause 20: The method of clause 19, wherein the loss comprises a chamfer distance between the estimated state of the channel and the actual state of the channel.

[0131] Clause 21: The method of clause 20, wherein the loss further comprises a weighted binary cross entropy between the estimated representation of the channel and the actual state of the channel.

[0132] Clause 22: A method according to any one of clauses 13 to 21, wherein training the machine learning model includes masking components in the estimated representation of the channel for components having confidence scores below a threshold confidence score.

[0133] Clause 23: A processing system comprising: a memory including computer executable instructions; and one or more processors configured to execute the computer executable instructions and cause the processing system to perform a method according to any one of clauses 1 to 22.

[0134] Clause 24: A processing system comprising: means for performing the method according to any one of clauses 1 to 22.

[0135] Clause 25: A non-transitory computer readable medium comprising computer executable instructions that, when executed by one or more processors of a processing system, cause the processing system to perform the method of any one of clauses 1 to 22.

[0136] Clause 26: A computer program product embodied on a computer readable storage medium, the computer readable storage medium comprising: code for performing the method according to any one of clauses 1 to 22.

[0137] Additional considerations

[0138] The previous description is provided to enable any person skilled in the art to practice the various embodiments described herein. The examples discussed herein are not limitations on the scope, applicability, or embodiments set forth in the claims. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the universal principles defined herein may be applied to other embodiments. For example, without departing from the scope of the present disclosure, the functions and arrangements of the elements discussed may be changed. Various examples may omit, replace, or add various procedures or components as appropriate. For example, the described methods may be performed in a sequence different from that described, and various steps may be added, omitted, or combined. In addition, the features described for some examples may be combined in some other examples. For example, any number of aspects set forth herein may be used to implement a device or practice method. In addition, the scope of the present disclosure is intended to cover such devices or methods practiced using other structures, functionality, or structures and functionality that are supplementary or alternative to the various aspects of the present disclosure set forth herein. It should be understood that any aspect of the disclosure disclosed herein may be embodied by one or more elements of the present invention.

[0139] As used herein, the word “exemplary” means “serving as an example, instance, or illustration.” Any aspect described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other aspects.

[0140] As used herein, a phrase referring to "at least one of" a list of items refers to any combination of those items (including single members). For example, "at least one of a, b, or c" is intended to cover a, b, c, ab, ac, bc, and abc, as well as any combination with multiple identical elements (e.g., aa, aaa, aab, aac, abb, acc, bb, bbb, bbc, cc, and ccc, or any other ordering of a, b, and c).

[0141] As used herein, the term "determining" encompasses a wide variety of actions. For example, "determining" may include calculating, computing, processing, deriving, investigating, searching (e.g., searching in a table, database, or other data structure), ascertaining, and the like. Furthermore, "determining" may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), and the like. Furthermore, "determining" may include resolving, selecting, choosing, establishing, and the like.

[0142] The method disclosed herein includes one or more steps or actions for implementing the method. The steps and / or actions of the method can be interchangeable with each other without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order and / or use of specific steps and / or actions can be modified without departing from the scope of the claims. In addition, the various operations of the method described above can be performed by any appropriate component capable of performing the corresponding function. The component may include various hardware and / or software components and / or modules, including but not limited to circuits, application specific integrated circuits (ASICs) or processors. Typically, in the case of operations illustrated in the accompanying drawings, those operations may have corresponding corresponding components with similar numbers plus functional components.

[0143] The following claims are not intended to be limited to the embodiments shown herein, but should be granted the full scope consistent with the language of the claims. Within the claims, unless otherwise specified, the reference to the singular element is not intended to mean "one and only one", but "one or more". Unless otherwise specified, the term "some" refers to one or more. No claim element should be interpreted according to the provisions of 35 U.S.C. § 112 (f) unless the element is explicitly recorded using the phrase "a component for...", or in the case of a method claim, the element is recorded using the phrase "a step for...". All structural and functional equivalents of the elements of the various aspects described throughout the present disclosure that are known or will be known later to a person of ordinary skill in the art are expressly incorporated herein by reference and are intended to be covered by the claims. In addition, nothing disclosed herein is intended to be dedicated to the public, regardless of whether such disclosure is explicitly recorded in the claims.

Claims

1. A computer-implemented method comprising: estimating a representation of a channel using a machine learning model trained to generate an estimated representation of the channel based on a location of a transmitter in a spatial environment, a location of a receiver in the spatial environment, and information about the spatial environment; as well as One or more actions are taken based on the estimated representation of the channel. 2 . The method of claim 1 , wherein the information about the spatial environment comprises a two-dimensional representation of a layout of the spatial environment.

3. The method of claim 1, wherein the machine learning model comprises: a first sub-model configured to encode the information about the spatial environment into an embedded representation of the spatial environment; and a second sub-model, the second sub-model being trained to generate the estimated representation of the channel as a continuous representation based on the embedded representation of the spatial environment, the position of the transmitter in the spatial environment, and the position of the receiver in the spatial environment.

4. The method according to claim 3, further comprising: generating a higher dimensional representation of the position of the transmitter in the spatial environment based on a Fourier transform; generating a higher dimensional representation of the position of the receiver in the spatial environment based on a Fourier transform; as well as The higher dimensional representation of the transmitter in the spatial environment and the higher dimensional representation of the position of the receiver in the spatial environment are input into the second sub-model.

5. The method of claim 3, wherein the first sub-model comprises a variational autoencoder.

6. The method of claim 1 , wherein the estimated representation of the channel comprises a plurality of vectors, each respective vector of the plurality of vectors corresponding to a respective signal component estimated to be received at the receiver, each respective vector comprising information identifying a received power, a flight time, a departure angle, and an arrival angle of the respective signal component associated with the respective vector.

7. The method of claim 1, wherein the machine learning model comprises a model trained to minimize a chamfer distance over time between an estimated state of the channel and an actual state of the channel.

8. The method of claim 1 , wherein the machine learning model comprises a model trained to minimize a loss between an estimated state of the channel and an actual state of the channel, the loss comprising a sum of a chamfer distance between the estimated state of the channel and the actual state of the channel and a weighted binary cross entropy.

9. The method of claim 1, wherein taking one or more actions comprises generating a graphical rendering of the estimated representation of the channel in a representation of the spatial environment.

10. The method of claim 1, wherein taking one or more actions comprises selecting one or more beams for communicating between the transmitter and the receiver based on the estimated representation of the channel.

11. The method of claim 1, wherein taking one or more actions comprises detecting a presence and a location of an object in the spatial environment based on the estimated representation of the channel.

12. The method of claim 1, wherein taking one or more actions comprises identifying material properties of the spatial environment based on the estimated representation of the channel.

13. A computer-implemented method comprising: generating a training data set comprising a plurality of samples, wherein each sample corresponds to a position of a receiver in a spatial environment and a position of a transmitter in the spatial environment, and each sample comprises a plurality of vectors, each respective vector representing a respective signal component received at the receiver at a particular position in the spatial environment; as well as A machine learning model is trained to generate an estimated representation of a channel based on information about the spatial environment and the training data set.

14. The method of claim 13, wherein training the machine learning model comprises: training a first sub-model to generate an embedded representation of the spatial environment based on the information about the spatial environment; as well as A second sub-model is trained based on the embedded representation of the spatial environment and the training data set to generate the estimated representation of the channel.

15. The method of claim 14, wherein the first sub-model comprises a variational autoencoder.

16. The method of claim 14, wherein the first sub-model is trained using an unsupervised learning technique and the second sub-model is trained using a supervised learning technique.

17. The method of claim 13, wherein each respective vector includes information identifying a received power, a time of flight, a departure angle, and an arrival angle of the respective signal component associated with the respective vector.

18. The method of claim 13, wherein: The location of the transmitter in the spatial environment comprises a high-dimensional representation of the location of the transmitter generated based on a Fourier transform applied to the three-dimensional location of the transmitter, and The position of the receiver in the spatial environment comprises a high-dimensional representation of the position of the receiver generated based on a Fourier transform applied to the three-dimensional position of the receiver.

19. The method of claim 13, wherein training the machine learning model comprises training the machine learning model to minimize a loss between an estimated state of the channel and an actual state of the channel.

20. The method of claim 19, wherein the loss comprises a chamfer distance between the estimated state of the channel and the actual state of the channel.

21. The method of claim 20, wherein the loss further comprises a weighted binary cross entropy between the estimated state of the channel and the actual state of the channel.

22. The method of claim 13, wherein training the machine learning model comprises masking components in the estimated representation of the channel for components having confidence scores below a threshold confidence score.

23. A processing system comprising: A memory, wherein executable instructions are stored in the memory; and a processor configured to execute the executable instructions so as to cause the processing system to: estimating a representation of a channel using a machine learning model trained to generate an estimated representation of the channel based on a location of a transmitter in a spatial environment, a location of a receiver in the spatial environment, and information about the spatial environment; as well as One or more actions are taken based on the estimated representation of the channel.

24. The processing system of claim 23, wherein the machine learning model comprises: a first sub-model configured to encode the information about the spatial environment into an embedded representation of the spatial environment; and a second sub-model, the second sub-model being trained to generate the estimated representation of the channel as a continuous representation based on the embedded representation of the spatial environment, the position of the transmitter in the spatial environment, and the position of the receiver in the spatial environment.

25. The processing system of claim 24, wherein the first sub-model comprises a variational autoencoder.

26. A processing system according to claim 23, wherein the estimated representation of the channel includes a plurality of vectors, each respective vector of the plurality of vectors corresponds to a respective signal component estimated to be received at the receiver, and each respective vector includes information identifying the received power, flight time, departure angle, and arrival angle of the respective signal component associated with the respective vector.

27. The processing system of claim 23, wherein the machine learning model comprises a model trained to minimize a chamfer distance over time between an estimated state of the channel and an actual state of the channel.

28. The processing system of claim 23, wherein the machine learning model comprises a model trained to minimize a loss over time between an estimated state of the channel and an actual state of the channel.

29. A processing system comprising: A memory, wherein executable instructions are stored in the memory; and a processor configured to execute the executable instructions so as to cause the processing system to: generating a training data set comprising a plurality of samples, wherein each sample corresponds to a position of a receiver in a spatial environment and a position of a transmitter in the spatial environment, and each sample comprises a plurality of vectors, each respective vector representing a respective signal component received at the receiver while traversing the spatial environment; as well as A machine learning model is trained to generate an estimated representation of a channel based on information about the spatial environment and the training data set.

30. The processing system of claim 29, wherein to train the machine learning model, the processor is configured to cause the processing system to: training a first sub-model to generate an embedded representation of the spatial environment based on the information about the spatial environment; and A second sub-model is trained based on the embedded representation of the spatial environment and the training data set to generate the estimated representation of the channel.

Images