Cyclic isomorphism reasoner for channel estimation

Abstract

Methods, systems, and devices are described for wireless communication. A wireless device can receive an assignment of a set of resources associated with a channel, where the set of resources includes a first subset of resources allocated for data transmissions and a second subset of resources allocated for reference signals. The wireless device can generate, from the reference signals, a first set of multiple channel estimates for each layer of the channel in accordance with a minimum mean square estimation (MMSE) operation. The wireless device can generate a second set of multiple channel estimates for each layer of the channel in accordance with a nonlinear two-dimensional interpolation of the channel, and can perform a refinement operation with the estimates to generate channel estimates associated with multiple layers.

Description

[0001] Cross Reference to Related Applications

[0002] This Patent Application claims priority to U.S. Patent Application No. 18 / 472,083 by PRATIK et al., entitled “RECURRENT EQUIVARIANT INFERENCE MACHINES FOR CHANNEL ESTIMATION,” filed September 21, 2023, and U.S. Patent Application No. 17 / 952,203 by PRATIK et al., entitled “RECURRENT EQUIVARIANT INFERENCE MACHINES FOR CHANNEL ESTIMATION,” filed September 23, 2022, each of which is assigned to the assignee hereof and expressly incorporated by reference herein in its entirety. BACKGROUND

[0003] The following relates to wireless communications, and more specifically to estimating a channel using a machine learning model.

[0004] Wireless communications systems are widely deployed to provide various types of communication content such as voice, video, packet data, messaging, broadcast, and so on. These systems can be capable of supporting communication with multiple users by sharing the available system resources (e.g., time, frequency, and power). Examples of such multiple- access systems include fourth generation (4G) systems such as Long Term Evolution (LTE) systems, LTE-Advanced (LTE-A) systems, or LTE-A Pro systems, and fifth generation (5G) systems which can be referred to as New Radio (NR) systems. These systems can employ technologies such as code division multiple access (CDMA), time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal FDMA (OFDMA), or discrete Fourier transform spread orthogonal frequency division multiplexing (DFT-S-OFDM). A wireless multiple-access communications system can include one or more base stations each simultaneously supporting communications with multiple communication devices, which can be otherwise known as user equipment (UE). SUMMARY

[0005] The described techniques relate to improved methods, systems, devices, and apparatuses that support cyclic equivariant inference machines for channel estimation. For example, the described techniques allow for the use of cyclic equivariant inference machines to compute channel estimates. In some cases, a wireless communication system can place pilot symbols (e.g., demodulation reference signal (DMRS) symbols) in a transmission slot according to a known pattern, allowing a wireless device to estimate unknown resources of a channel based on known resources (e.g., DMRS symbols). For example, a wireless device can receive an assignment of a set of resources associated with a channel, where the set of resources includes a first subset of resources allocated for data transmission and a second subset of resources allocated for reference signals (e.g., DMRS). The wireless device can generate a plurality of channel estimates (e.g., single-input and single-output (SISO) channel estimates) for each layer of the channel, and utilize the estimates to perform a refinement operation to generate a channel estimate associated with multiple layers (e.g., multiple-input and multiple-output (MIMO) channel estimate). In some cases, the refinement operation can include a plurality of iterations. For example, each iteration can include generating a respective gradient associated with each of the per-layer channel estimates based on the per-layer channel estimates and observed resources of the channel (e.g., known DMRS resources); generating a current set of values of a latent variable (e.g., an inference variable based on an observed variable) based on a previous set of values of the latent variable, the respective gradient, and the per-layer channel estimates; and modifying (e.g., refining, updating, improving) the channel estimate based on the current set of values of the latent variable, the per-layer channel estimates, and the respective gradient. In some cases, the refinement operation can be performed by a refinement network that includes a likelihood module, an encoder module, and a decoder module.

[0006] A method for wireless communication at a wireless communication device is described. The method can include receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal, generating, from the reference signal received on the second subset of resources, a first set of channel estimates associated with respective layers of a set of layers of the channel for the set of resources according to a minimum mean square estimation (MMSE) operation, generating a second set of channel estimates and a set of values of a latent variable associated with respective layers of the set of layers of the channel for the set of resources according to a nonlinear two-dimensional interpolation of the channel, the nonlinear two-dimensional interpolation of the channel based on the first set of channel estimates, and including a refinement operation that includes one or more iterations on the second set of channel estimates, where each iteration of the one or more iterations is performed according to a same set of machine learning parameters. Performing each iteration of the one or more iterations can include generating a respective gradient associated with the second set of channel estimates based on the second set of channel estimates for the second subset of resources and a measured observation of the second subset of resources, generating a second set of values of the latent variable based on a first set of values of the set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient, and modifying the second set of channel estimates associated with the set of layers based on the second set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient.

[0007] An apparatus for wireless communication at a wireless communication device is described. The apparatus can include one or more processors and one or more memories coupled with the one or more processors. The one or more processors can be configured to cause the wireless communication device to receive an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal, generate, from the reference signal received on the second subset of resources, a first set of channel estimates associated with respective layers of a set of layers of the channel for the set of resources according to MMSE operations, generate a second set of channel estimates and a set of values of a latent variable associated with respective layers of the set of layers of the channel for the set of resources according to a nonlinear two-dimensional interpolation of the channel for the set of resources, where the nonlinear two-dimensional interpolation of the channel is based on the first set of channel estimates, and perform a refinement operation including one or more iterations on the second set of channel estimates, where each iteration of the one or more iterations is performed according to a same set of machine learning parameters. To perform each iteration of the one or more iterations, the one or more processors can be configured to cause the wireless device to generate a respective gradient associated with the second set of channel estimates based on the second set of channel estimates for the second subset of resources and a measured observation of the second subset of resources, generate a second set of values of the latent variable based on a first set of values of the set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient, and modify the second set of channel estimates associated with the set of layers based on the second set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient.

[0008] Another apparatus for wireless communication is described. The apparatus can include means for receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal, generating, from the reference signal received on the second subset of resources, a first set of channel estimates associated with respective layers of a set of layers of the channel for the set of resources according to MMSE operation, generating a second set of channel estimates and a set of values of a latent variable for respective layers of the set of layers of the channel for the set of resources according to a nonlinear two-dimensional interpolation of the channel, the nonlinear two-dimensional interpolation of the channel based on the first set of channel estimates, and including a refinement operation of the second set of channel estimates comprising one or more iterations, where each iteration of the one or more iterations is performed according to a same set of machine learning parameters. Performing each iteration of the one or more iterations can include means for generating a respective gradient associated with the second set of channel estimates based on the second set of channel estimates for the second subset of resources and a measurement of the second subset of resources, generating a second set of values of the latent variable based on a first set of values of the set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient, and modifying the second set of channel estimates associated with the set of layers based on the second set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient.

[0009] A non-transitory computer-readable medium storing code for wireless communication at a wireless communication device is described. The code can include instructions executable by one or more processors to cause the wireless communication device to receive an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal, generate, from the reference signal received on the second subset of resources, a first set of channel estimates associated with respective layers of a set of layers of the channel for the set of resources according to MMSE operations, generate a second set of channel estimates and a set of values of a latent variable associated with respective layers of the set of layers of the channel for the set of resources according to a nonlinear two-dimensional interpolation of the channel for the set of resources, where the nonlinear two-dimensional interpolation of the channel is based on the first set of channel estimates, and perform a refinement operation including one or more iterations on the second set of channel estimates, where each iteration of the one or more iterations is performed according to a same set of machine learning parameters. The instructions to perform each iteration of the one or more iterations can be executable by the one or more processors to cause the wireless device to generate a respective gradient associated with the second set of channel estimates based on the second set of channel estimates for the second subset of resources and an observed measurement of the second subset of resources, generate a second set of values of the latent variable based on a first set of values of the set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient, and modify the second set of channel estimates associated with the set of layers based on the second set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient.

[0010] In some examples of the method, apparatuses, and non-transitory computer- readable medium described herein, the refinement operation is a first refinement operation and the set of machine learning parameters is a first set of machine learning parameters. Some examples of the method, apparatuses, and non-transitory computer-readable medium described herein can include operations, features, means, or instructions for performing a second refinement operation on the second set of channel estimates, the second refinement operation including one or more second iterations performed according to a same second set of machine learning parameters, where the first refinement operation and the second refinement operation are associated with respective attention computations of the set of attention computations.

[0011] In some examples of the method, apparatuses, and non-transitory computer- readable medium described herein, the set of multiple attention computations includes intra-physical resource block (PRB) group computations, inter-PRB group computations, cross multiple-input multiple-output (MIMO) computations, multi-layer perceptron (MLP) computations, or any combination thereof.

[0012] Some examples of the method, apparatuses, and non-transitory computer- readable medium described herein can include operations, features, means, or instructions for performing the MMSE operation based on a resource configuration mode of the second subset of resources allocated for the reference signal, the reference signal including a demodulation reference signal (DMRS).

[0013] In some examples of the method, apparatuses, and non-transitory computer- readable medium described herein, generating the respective gradient can include operations, features, means, or instructions for generating a respective set of values of a residual variable based on a difference between an observation of the measurement of the second subset of resources and the second set of multiple channel estimates of the second subset of resources, and combining the respective set of values of the residual variable, the second subset of resources, and a number of mask bits.

[0014] In some examples of the method, apparatuses, and non-transitory computer- readable medium described herein, generating the second set of values of the latent variable can include operations, features, means, or instructions for combining, based on generating the respective gradient, the second set of multiple channel estimates for the second subset of resources, the respective gradient, and a respective value of the first set of values of the latent variable.

[0015] In some examples of the method, apparatuses, and non-transitory computer- readable medium described herein, generating the second set of values of the latent variable can include operations, features, means, or instructions for modeling a correlation between resources of each resource block in a resource block group and other resource blocks in the resource block group.

[0016] In some examples of the method, apparatuses, and non-transitory computer- readable medium described herein, generating the second set of values of the latent variable can include operations, features, means, or instructions for modeling a correlation between resources of each group in a set of multiple groups of resources of the resource set and other groups in the set of multiple groups of resources, where each group in the set of multiple groups of resources includes a set of multiple resource blocks.

[0017] In some examples of the method, apparatuses, and non-transitory computer- readable medium described herein, generating the second set of values of the latent variable can include operations, features, means, or instructions for modeling a correlation between each layer in the set of multiple layers of the resource set.

[0018] In some examples of the method, apparatus, and non-transitory computer-readable medium described herein, modifying the second set of multiple channel estimates can include operations, features, means, or instructions for combining the second set of values of the latent variable, the second set of multiple channel estimates, and the respective gradient based on the set of machine learning parameters.

[0019] In some examples of the method, apparatus, and non-transitory computer-readable medium described herein, the non-linear two-dimensional interpolation of the channel is based on a machine learning model.

[0020] In some examples of the method, apparatus, and non-transitory computer-readable medium described herein, the first set of multiple channel estimates and the second set of multiple channel estimates are associated with a set of multiple single-input single-output (SISO) antenna pairs.

[0021] In some examples of the method, apparatus, and non-transitory computer-readable medium described herein, each iteration of the one or more iterations is performed by a refinement network including a likelihood module, an encoder module, and a decoder module, the refinement network including a machine learning model. In some examples of the method, apparatus, and non-transitory computer-readable medium described herein, each refinement network is performed according to the same set of machine learning parameters.

[0022] A method for wireless communication is described. The method can include receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal, generating a set of multiple channel estimates associated with respective layers of a set of multiple layers of the channel for the set of resources, and performing a refinement operation including one or more iterations on the set of multiple channel estimates, where each iteration of the one or more iterations can include operations, features, means, or instructions for generating a respective gradient associated with the set of multiple channel estimates based on a second set of multiple channel estimates for the second subset of resources and an observation of a measurement of the second subset of resources, generating a second set of values of a latent variable based on a first set of values of the latent variable, the set of multiple channel estimates, and the respective gradient, and modifying the set of multiple channel estimates associated with the set of multiple layers based on the second set of values of the latent variable, the set of multiple channel estimates, and the respective gradient.

[0023] An apparatus for wireless communication at a wireless communication device is described. The apparatus can include one or more processors and one or more memories coupled with the one or more processors. The one or more processors can be configured to cause the wireless communication device to receive an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal, generate a set of multiple channel estimates associated with respective layers of a set of multiple layers of the channel for the set of resources, and perform a refinement operation including one or more iterations on the set of multiple channel estimates, where the instructions of each iteration of the one or more iterations are executable by the processor to cause the apparatus to generate respective gradients associated with the set of multiple channel estimates based on a second set of multiple channel estimates for the second subset of resources and observations of measurements of the second subset of resources, generate a second set of values of a latent variable based on a first set of values of the latent variable, the set of multiple channel estimates, and the respective gradients, and modify the set of multiple channel estimates associated with the set of multiple layers based on the second set of values of the latent variable, the set of multiple channel estimates, and the respective gradients.

[0024] Another apparatus for wireless communication is described. The apparatus can include means for receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal, means for generating a set of multiple channel estimates associated with respective layers of a set of multiple layers of the channel for the set of resources, and means for performing a refinement operation including one or more iterations on the set of multiple channel estimates, where the means for each iteration of the one or more iterations can include means for generating respective gradients associated with the set of multiple channel estimates based on a second set of multiple channel estimates for the second subset of resources and observations of measurements of the second subset of resources, means for generating a second set of values of a latent variable based on a first set of values of the latent variable, the set of multiple channel estimates, and the respective gradients, and means for modifying the set of multiple channel estimates associated with the set of multiple layers based on the second set of values of the latent variable, the set of multiple channel estimates, and the respective gradients.

[0025] A non-transitory computer-readable medium storing code for wireless communication is described. The code can include instructions executable by one or more processors to receive an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal, generate a set of multiple channel estimates associated with respective layers of a set of multiple layers of the channel for the set of resources, and perform a refinement operation including one or more iterations on the set of multiple channel estimates, where the instructions of each iteration of the one or more iterations are executable to generate a respective gradient associated with the set of multiple channel estimates based on a second set of multiple channel estimates for the second subset of resources and an observed measurement of the second subset of resources, generate a second set of values of a latent variable based on a first set of values of the latent variable, the set of multiple channel estimates, and the respective gradient, and modify the set of multiple channel estimates associated with the set of multiple layers based on the second set of values of the latent variable, the set of multiple channel estimates, and the respective gradient.

[0026] In some examples of the method, apparatuses, and non-transitory computer- readable medium described herein, generating the respective gradient can include operations, features, means, or instructions for generating a respective set of values of a residual variable based on a difference between the observed measurement of the second subset of resources and the set of multiple channel estimates of the second subset of resources, and combining the respective set of values of the residual variable, the observed measurement of the second subset of resources, and a number of mask bits.

[0027] In some examples of the method, apparatuses, and non-transitory computer- readable medium described herein, generating the second set of values of the latent variable can include operations, features, means, or instructions for combining, based on generating the respective gradient, the set of multiple channel estimates for the second subset of resources, the respective gradient, and a respective value of the first set of values of the latent variable.

[0028] In some examples of the method, apparatuses, and non-transitory computer- readable medium described herein, generating the second set of values of the latent variable can include operations, features, means, or instructions for modeling a correlation between a resource of each resource block in a resource block group and other resource blocks in the resource block group.

[0029] In some examples of the method, apparatus, and non-transitory computer-readable medium described herein, generating the second set of values of the latent variable can include operations, features, means, or instructions for modeling a correlation between resources of each group of a set of multiple groups of resources of the set of resources and other groups of the set of multiple groups of resources, where each group of the set of multiple groups of resources includes a set of multiple resource blocks.

[0030] In some examples of the method, apparatus, and non-transitory computer-readable medium described herein, generating the second set of values of the latent variable can include operations, features, means, or instructions for modeling a correlation between each layer of the set of multiple layers of the set of resources.

[0031] In some examples of the method, apparatus, and non-transitory computer- readable medium described herein, modifying the set of multiple channel estimates can include operations, features, means, or instructions for combining the second set of values of the latent variable, the set of multiple channel estimates, and the respective gradient.

[0032] In some examples of the method, apparatus, and non-transitory computer- readable medium described herein, an initial value of the set of multiple channel estimates can be associated with a SISO antenna pair.

[0033] In some examples of the method, apparatus, and non-transitory computer- readable medium described herein, the second subset of resources can be configured according to a resource configuration pattern of a set of resource configuration patterns.

[0034] In some examples of the method, apparatus, and non-transitory computer- readable medium described herein, the set of resource configuration patterns can be a set of DMRS patterns.

[0035] In some examples of the method, apparatus, and non-transitory computer- readable medium described herein, each iteration can be performed by a refinement network including a likelihood module, an encoder module, and a decoder module, and each refinement network further includes respective parameters associated with machine learning operations.

[0036] In some examples of the method, apparatus, and non-transitory computer- readable medium described herein, the set of resources includes one or more groups of resources, and each respective layer of the set of multiple layers can be associated with a respective antenna pair of a set of multiple SISO antenna pairs. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 An example of a wireless communications system that supports recurrent equivariant reasoning machines for channel estimation in accordance with one or more aspects of the present disclosure is shown.

[0038] Figure 2A diagram illustrating an example of a network that supports cyclic equivariant inference machine for channel estimation in accordance with one or more aspects of the present disclosure is shown.

[0039] Figures 3 to 6 A diagram illustrating an example of a network that supports cyclic equivariant inference machine for channel estimation in accordance with one or more aspects of the present disclosure is shown.

[0040] Figure 7 And Figure 8 A block diagram of a device that supports cyclic equivariant inference machine for channel estimation in accordance with one or more aspects of the present disclosure is shown.

[0041] Figure 9 A block diagram of a communications manager that supports cyclic equivariant inference machine for channel estimation in accordance with one or more aspects of the present disclosure is shown.

[0042] Figure 10 A diagram of a system including a user equipment (UE) that supports cyclic equivariant inference machine for channel estimation in accordance with one or more aspects of the present disclosure is shown.

[0043] Figure 11 A diagram of a system including a network entity that supports cyclic equivariant inference machine for channel estimation in accordance with one or more aspects of the present disclosure is shown.

[0044] Figures 12 to 15 A flow diagram illustrating a method that supports cyclic equivariant inference machine for channel estimation in accordance with one or more aspects of the present disclosure is shown. DETAILED DESCRIPTION

[0045] Some wireless communication systems can support channel estimation. For example, a wireless device can estimate resources of a channel to maintain high data throughput. To enable channel estimation, some communication systems can utilize a tracking reference signal (TRS) to compute channel properties (e.g., Doppler, delay spread, signal-to-noise ratio (SNR), etc.) and estimate channel resources based on the channel properties. However, TRSs can include a relatively large overhead (e.g., memory, computation), and not all wireless communication systems can transmit TRSs continuously (e.g., periodically or relatively regularly over time), which can impact channel estimation. Additionally, some wireless communication systems can support multiple-input and multiple-output (MIMO) communications, where multiple resource layers can interfere with each other, further reducing accuracy of channel estimation. A resource layer can refer to a spatial layer of a wireless channel. For MIMO communications, multiple antennas or antenna ports of a transmitting device can each be associated with a respective layer, which can transmit on the same time-frequency resource. To further improve TRS consistency, and considering cross-MIMO interference, a channel estimation process can be updated.

[0046] Some wireless communication devices support machine learning based channel estimation operations. The techniques described herein allow the techniques of minimum mean square estimation (MMSE) operations for channel estimation to be applied to machine learning based channel estimation techniques to improve the channel estimation techniques. For example, performing a combination of machine learning based and MMSE based channel estimation can provide enhanced channel estimation through machine learning capabilities while maintaining existing channel estimation and de-mapping hardware at the wireless device for MMSE operations, which can reduce memory consumption and computational costs among other possibilities. MMSE operations can include a wireless device estimating a channel by minimizing the mean square error of a variable associated with the channel. The techniques described herein can allow the use of a cyclic equivariant inference machine to compute channel estimates, which can be a type of machine learning model that provides relatively reliable and accurate estimates based on a given set of inputs. In some cases, a wireless communication system can place pilot symbols (e.g., demodulation reference signal (DMRS) symbols, other types of reference symbols, or any other pilot symbols) in a transmission slot according to a known pattern, allowing a wireless device to estimate unknown resources of a channel based on known resources (e.g., DMRS symbols). For example, a wireless device can receive an assignment of a first set of resources within a channel allocated for data transmission and a second set of resources within the channel allocated for reference signals (e.g., DMRS). The wireless device can perform MMSE operations to generate a first set of multiple channel estimates per layer of the channel (e.g., single input and single output (SISO) channel estimates). MMSE operations can be an estimation technique that utilizes linear equalization to estimate a channel. MMSE can be supported by hardware within the wireless device, which can enable reduced complexity and processing compared to performing an initial estimate based on machine learning. The wireless device can then utilize non-linear two-dimensional interpolation to modify the MMSE estimates and generate a second set of multiple channel estimates per layer of the channel. For example, the wireless device can construct the first set of multiple channel estimates using interpolation in time and frequency domains according to a machine learning model. In some cases, the non-linear two-dimensional interpolation can utilize a machine learning model to improve the accuracy of the channel estimates produced by the MMSE operations and apply the channel estimates to time and frequency domains.

[0047] The wireless device can utilize the second set of channel estimates to perform a refinement operation to generate a channel estimate associated with the plurality of layers (e.g., a MIMO channel estimate). That is, in one example, the wireless device can take the channel estimates generated via the MMSE operation and subsequent machine learning interpolation, and the wireless device can refine the channel estimates and combine the estimates of each layer of the channel into a single MIMO channel estimate. In some cases, the refinement operation can include multiple iterations of refinement. Each refinement iteration can generate a respective gradient associated with each layer channel estimate based on the observed resources of the channel (e.g., known DMRS resources) and each layer channel estimate. Each refinement iteration can further generate a current set of values for latent variables (e.g., inferred variables based on the observed variables). The current set of values for the latent variables can be generated based on a previous set of values for the latent variables, the respective gradient, and each layer channel estimate. Each refinement iteration can further modify (e.g., refine, update, improve) the channel estimates based on the current set of values for the latent variables, each layer channel estimate, and the respective gradient. That is, in one example, each refinement iteration can further improve the channel estimates and generate new or improved values for the inferred latent variables. In some cases, the refinement operation can be performed by a refinement network, which can be a machine learning model that includes a likelihood module, an encoder module, and a decoder module.

[0048] The techniques described herein can allow a wireless device to reliably and accurately estimate a channel with relatively low memory consumption and processing. For example, by first utilizing an MMSE operation to generate channel estimates, the wireless device can reduce memory consumption and processing as compared to machine learning based estimates or other techniques. Additionally or alternatively, by utilizing MMSE for channel estimation, the wireless device can support reduced hardware cost and complexity as compared to other machine learning based channel estimation techniques, at least because the MMSE operation can be supported by current hardware components of the wireless device.

[0049] Aspects of the disclosure are first described in the context of a wireless communication system. Then, aspects of the disclosure are described in the context of a network. Aspects of the disclosure are further illustrated by and described in conjunction with apparatus diagrams, system diagrams, and flowcharts related to a cyclic equivariant inference machine for channel estimation.

[0050] Figure 1 An example of a wireless communication system 100 that supports a cyclic equivariant inference machine for channel estimation is shown in accordance with one or more aspects of the present disclosure. The wireless communication system 100 can include one or more network entities 105, one or more UEs 115, and a core network 130. In some examples, the wireless communication system 100 can be a LTE network, a LTE-A network, a LTE-A Pro network, a NR network, or a network operating according to other systems and radio technologies, including future systems and radio technologies not explicitly mentioned herein.

[0051] Network entities 105 can be dispersed throughout the geographic region of wireless communication system 100, and can each include devices of different form factors or having different capabilities. In various examples, network entities 105 can be referred to as network elements, mobility elements, radio access network (RAN) nodes, or network equipment, among other nomenclature. In some examples, network entities 105 and UEs 115 can wirelessly communicate via one or more communication links 125 (e.g., radio frequency (RF) access links). For example, a network entity 105 can support a coverage area 110 (e.g., a geographic coverage area) within which UEs 115 and network entities 105 can establish one or more communication links 125. Coverage area 110 can be an example of a geographic area over which network entities 105 and UEs 115 can support signal communication in accordance with one or more radio access technologies (RATs).

[0052] UEs 115 can be dispersed throughout the coverage areas 110 of wireless communication system 100, and each UE 115 can be stationary or mobile, or at different times be stationary and mobile. UEs 115 can be devices of different form factors Figure 1 Some example UEs 115 are illustrated. UEs 115 as described herein can be able to communicate with various types of devices, such as other UEs 115 or network entities 105 as discussed Figure 1 As illustrated, a UE 115 can include a communication manager 101 configured to transmit communications to and receive communications from a network entity 105. In some examples, the communication manager 101 can be configured to receive an assignment of a set of resources, the set of resources including resources allocated for a data signal and resources allocated for a reference signal. Additionally or alternatively, the communication manager 101 can be configured to generate a channel estimate based on MMSE operation, non-linear two-dimensional interpolation of a channel, a refinement operation, or any combination thereof.

[0053] As described herein, a node, which can be referred to as a network node, network entity, or wireless node, can be a base station (e.g., any of the base stations described herein), a UE (e.g., any of the UEs described herein), a network controller, an apparatus, a device, a computing system, one or more components, and / or another suitable processing entity configured to perform any of the techniques described herein. For example, a network node can be a UE. As another example, a network node can be a base station. As yet another example, a first network node can be configured to communicate with a second network node or a third network node. In one aspect of this example, the first network node can be a UE, the second network node can be a base station, and the third network node can be a UE. In another aspect of this example, the first network node can be a UE, the second network node can be a base station, and the third network node can be a base station. In yet other aspects of this example, the first network node, the second network node, and the third network node can be different relative to these examples. Similarly, a reference to a UE, a base station, an apparatus, a device, a computing system, etc., can include a disclosure of a UE, a base station, an apparatus, a device, a computing system, etc., as a network node. For example, a disclosure that a UE is configured to receive information from a base station also discloses that a first network node is configured to receive information from a second network node. Once a particular example has been expanded in accordance with the present disclosure (e.g., a disclosure that a UE is configured to receive information from a base station also discloses that a first network node is configured to receive information from a second network node), the narrower example can be interpreted in reverse, but in a broad, open-ended fashion. In the above example in which a disclosure that a UE is configured to receive information from a base station also discloses that a first network node is configured to receive information from a second network node, the first network node can refer to a first UE, a first base station, a first apparatus, a first device, a first computing system, a first one or more components, a first processing entity, etc., configured to receive the information; and the second network node can refer to a second UE, a second base station, a second apparatus, a second device, a second computing system, a second one or more components, a second processing entity, etc.

[0054] As described herein, different terminology can be used in various aspects to describe the communication of information (e.g., any information, signals, etc.). A disclosure using one communication term includes a disclosure using other communication terms. For example, a first network node can be described as being configured to send information to a second network node. In this example and consistent with the present disclosure, a disclosure that the first network node is configured to send information to the second network node includes a disclosure that the first network node is configured to provide, transmit, output, communicate, or send information to the second network node. Similarly, in this example and consistent with the present disclosure, a disclosure that the first network node is configured to send information to the second network node includes a disclosure that the second network node is configured to receive, obtain, or decode the information provided, transmitted, output, communicated, or sent by the first network node.

[0055] In some examples, the network entities 105 can communicate with the core network 130, or with each other, or both. For example, the network entities 105 can communicate with the core network 130 via one or more backhaul communication links 120 (e.g., according to an SI, N2, N3, or other interface protocol). In some examples, the network entities 105 can communicate with each other via the backhaul communication links 120 (e.g., according to an X2, Xn, or other interface protocol) either directly (e.g., direct point-to-point between network entities 105) or indirectly (e.g., via core network 130). In some examples, the network entities 105 can communicate with each other via mid-cell communication links 162 (e.g., according to a mid-cell interface protocol) or front-haul communication links 168 (e.g., according to a front-haul interface protocol), or any combination thereof. The backhaul communication links 120, mid-cell communication links 162, or front-haul communication links 168 can be or include one or more wired links (e.g., electrical, fiber optic), one or more wireless links (e.g., radio, wireless optical), etc., or various combinations thereof. A UE 115 can communicate with the core network 130 via communication links 155.

[0056] One or more of the network entities 105 described herein can include or can be referred to as a base station 140 (e.g., a transceiver base station, a radio base station, an NR base station, an access point, a radio transceiver, a NodeB, an eNodeB (eNB), a next-generation NodeB, or a giga-NodeB (either of which can be referred to as a gNB), a 5G NB, a next-generation eNB (ng-eNB), a Home NodeB, a Home eNodeB, or other suitable terminology). In some examples, the network entities 105 (e.g., base stations 140) can be implemented in an aggregated (e.g., monolithic, self-standing) base station architecture that can be configured to utilize protocol stacks that are physically or logically integrated within a single network entity 105 (e.g., a single RAN node such as a base station 140). The network entities 105 can include a communication manager 102 configured to transmit communications to and receive communications from a UE 115. In some examples, the communication manager 102 can be configured to transmit an assignment of a set of resources including resources allocated for a data signal and resources allocated for a reference signal. Additionally or alternatively, the communication manager 102 can be configured to generate a channel estimate based on MMSE operation, non-linear two-dimensional interpolation of a channel, a refinement operation, or any combination thereof.

[0057] The techniques described herein can be implemented via additional or alternative wireless devices, including IAB nodes 104, distributed units (DUs) 165, centralized units (CUs) 160, radio units (RUs) 170, etc., in addition to or instead of being implemented between a UE 115 and a network entity 105. For example, in some implementations, aspects described herein can be implemented in the context of a disaggregated radio access network (RAN) architecture (e.g., an open RAN architecture). In a disaggregated architecture, a RAN can be split into three functional areas corresponding to the CU 160, the DU 165, and the RU 170. The functional split between the CU 160, the DU 165, and the RU 170 is flexible, and thus gives rise to many permutations of functionality depending on which functions are performed at the CU 160, the DU 165, and the RU 170 (e.g., MAC functions, baseband functions, radio frequency functions, and any combination thereof). For example, a functional split of the protocol stack can be employed between the DU 165 and the RU 170, such that the DU 165 can support one or more layers of the protocol stack, and the RU 170 can support one or more different layers of the protocol stack.

[0058] Some wireless communications systems (e.g., wireless communications system 100), infrastructure for NR access, and spectrum resources can additionally support wireless backhaul link capabilities to supplement wired backhaul connections, providing an IAB network architecture. One or more network entities 105 can include a CU 160, a DU 165, and a RU 170, and can be referred to as a donor network entity 105 or an IAB donor. One or more DUs 165 (e.g., and / or RUs 170) associated with a donor network entity 105 can be partially controlled by a CU 160 associated with the donor network entity 105. One or more donor network entities 105 (e.g., IAB donors) can communicate with one or more additional network entities 105 (e.g., IAB nodes 104) via supported access and backhaul links. An IAB node 104 can support mobile terminal (MT) functionality controlled and / or scheduled by a coupled IAB donor’s DU 165. Additionally, an IAB node 104 can include a DU 165 that supports communication links with additional entities (e.g., IAB nodes 104, UEs 115, etc.) within a relay chain or configuration (e.g., downstream) of an access network. In such cases, one or more components of a disaggregated RAN architecture (e.g., one or more IAB nodes 104 or components of an IAB node 104) can be configured to operate according to the techniques described herein.

[0059] In some examples, the wireless communications system 100 can include a core network 130 (e.g., a Next Generation Core (NGC)), one or more IAB donors, IAB nodes 104, and UEs 115, where the IAB nodes 104 can be controlled in part by each other and / or the IAB donors. The IAB donor nodes and the IAB nodes 104 can be examples of aspects of network entity 105. The IAB donor and the one or more IAB nodes 104 can be configured as (or communicate according to) a certain relay chain.

[0060] For example, an access network (AN) or RAN can refer to communications between an access node (e.g., an IAB donor), IAB nodes 104, and one or more UEs 115. An IAB donor can facilitate a connection between the core network 130 and the AN (e.g., via a wired or wireless connection to the core network 130). That is, an IAB donor can refer to a RAN node that has a wired or wireless connection to the core network 130. An IAB donor can include a CU 160 and at least one DU 165 (e.g., and RU 170), where the CU 160 can communicate with the core network 130 over an NG interface (e.g., some backhaul link). The CU 160 can host layer 3 (L3) (e.g., radio resource control (RRC), service data adaptation protocol (SDAP), PDCP, etc.) functionality and signaling. The at least one DU 165 and / or RU 170 can host lower layers, such as layer 1 (LI) and layer 2 (L2) (e.g., RLC, MAC, physical (PHY), etc.) functionality and signaling, and can each be controlled at least in part by the CU 160. The DU 165 can support one or more different cells. The IAB donor and the IAB nodes 104 can communicate over an Fl interface according to a certain protocol that defines signaling messages (e.g., an Fl AP protocol). Additionally, the CU 160 can communicate with the core network over an NG interface, which can be an example of a portion of a backhaul link, and can communicate with other CUs 160 (e.g., associated with an alternative IAB donor) over an Xn-C interface, which can be an example of a portion of a backhaul link.

[0061] An IAB node 104 can refer to a RAN node that provides IAB functionality (e.g., access for UEs 115, wireless self-backhauling capabilities, etc.). An IAB node 104 can include a DU 165 and an MT. The DU 165 can act as a distributed scheduling node toward child nodes associated with the IAB node 104, and the MT can act as a scheduled node toward a parent node associated with the IAB node 104. That is, an IAB donor can be referred to as a parent node in communication with one or more child nodes (e.g., the IAB donor can relay transmissions of a UE through one or more other IAB nodes 104). Additionally, depending on the relay chain or configuration of ANs, an IAB node 104 can also be referred to as a parent node or a child node of other IAB nodes 104. Thus, the MT entity of an IAB node 104 (e.g., MT) can provide a Uu interface for a child node to receive signaling from a parent IAB node 104, and the DU interface (e.g., DU 165) can provide a Uu interface for a parent node to signal to a child IAB node 104 or a UE 115.

[0062] For example, an IAB node 104 can be referred to as a parent node associated with an IAB node and a child node associated with an IAB donor. An IAB donor can include a CU 160 with a wired (e.g., fiber) or wireless connection to a core network, and can act as a parent node of an IAB node 104. For example, a DU 165 of an IAB donor can relay transmissions to a UE 115 through an IAB node 104, and can signal transmissions directly to the UE 115. A CU 160 of an IAB donor can signal a communication link establishment to an IAB node 104 via an Fl interface, and the IAB node 104 can schedule transmissions (e.g., transmissions relayed from the IAB donor to the UE 115) through the DU 165. That is, data can be relayed to and from an IAB node 104 via signaling over an NR Uu interface to an MT of the IAB node 104. Communications with the IAB node 104 can be scheduled by a DU 165 of an IAB donor, and communications with the IAB node 104 can be scheduled by a DU 165 of the IAB node 104.

[0063] Where the techniques described herein are applied in the context of a disaggregated RAN architecture, one or more components of the disaggregated RAN architecture (e.g., one or more IAB nodes 104 or components of an IAB node 104) can be configured to support techniques for large round trip times in random access channel procedures, as described herein. For example, some operations described as being performed by a UE 115 or a network entity 105 can additionally or alternatively be performed by a component of a disaggregated RAN architecture (e.g., an IAB node, a DU, a CU, etc.).

[0064] In some examples, the network entity 105 can be implemented in a disaggregated architecture (e.g., a disaggregated base station architecture, a disaggregated RAN architecture) that can be configured to utilize a protocol stack that is physically or logically distributed among two or more network entities 105, such as an IAB network, an Open RAN (O-RAN) (e.g., a network configuration sponsored by the O-RAN Alliance), or a virtualized RAN (vRAN) (e.g., a cloud RAN (C-RAN)). For example, the network entity 105 can include one or more of a CU 160, a DU 165, a RU 170, a RAN intelligent controller (RIC) 175 (e.g., a near real-time RIC (near-RT RIC), a non-real-time RIC (non-RT RIC)), a service management and orchestration (SMO) system 180, or any combination thereof. The RU 170 can also be referred to as a radio head, an intelligent radio head, a remote radio head (RRH), a remote radio unit (RRU), or a transmission reception point (TRP). One or more components of the network entity 105 in a disaggregated RAN architecture can be co-located, or one or more components of the network entity 105 can be in distributed locations (e.g., separate physical locations). In some examples, one or more network entities 105 of a disaggregated RAN architecture can be implemented as virtual units (e.g., a virtual CU (VCU), a virtual DU (VDU), a virtual RU (VRU)).

[0065] The functional split between the CU 160, the DU 165, and the RU 170 is flexible and can support different functionality depending on which functions (e.g., network layer functions, protocol layer functions, baseband functions, RF functions, and any combination thereof) are performed at the CU 160, the DU 165, or the RU 170. For example, a functional split of a protocol stack can be employed between the CU 160 and the DU 165, such that the CU 160 can support one or more layers of the protocol stack and the DU 165 can support one or more different layers of the protocol stack. In some examples, the CU 160 can host higher protocol layer (e.g., Layer 3 (L3), Layer 2 (L2)) functionality and signaling (e.g., RRC, SDAP, PDCP). The CU 160 can be connected to one or more DUs 165 or RUs 170, and the one or more DUs 165 or RUs 170 can host lower protocol layers, such as Layer 1 (LI) (e.g., PHY layer) or L2 (e.g., RLC layer, Medium Access Control (MAC) layer) functionality and signaling, and can each be at least partially controlled by the CU 160. Additionally or alternatively, a functional split of a protocol stack can be employed between the DU 165 and the RU 170, such that the DU 165 can support one or more layers of the protocol stack and the RU 170 can support one or more different layers of the protocol stack. The DU 165 can support one or more different cells (e.g., via one or more RUs 170). In some cases, the functional split between the CU 160 and the DU 165 or between the DU 165 and the RU 170 can be within a protocol layer (e.g., some functions of a protocol layer can be performed by one of the CU 160, the DU 165, or the RU 170, while other functions of that protocol layer are performed by a different one of the CU 160, the DU 165, or the RU 170). The CU 160 can be further split in functionality into CU control plane (CU-CP) and CU user plane (CU-UP) functions. The CU 160 can be connected to one or more DUs 165 via a midhaul communication link 162 (e.g., Fl, Fl-c, Fl-u), and the DU 165 can be connected to one or more RUs 170 via a front-haul communication link 168 (e.g., open front-haul (FH) interface). In some examples, the midhaul communication link 162 or the front-haul communication link 168 can be implemented according to an interface (e.g., channel) between layers of a protocol stack that are supported by the respective network entities 105 that communicate via these communication links.

[0066] In some wireless communications systems (e.g., wireless communications system 100), infrastructure and spectrum resources for radio access can support wireless backhaul link capabilities to supplement wired backhaul connections to provide an IAB network architecture (e.g., to core network 130). In some cases, in an IAB network, one or more network entities 105 (e.g., IAB nodes 104) can be partially controlled by one another. One or more IAB nodes 104 can be referred to as a donor entity or IAB donor. One or more DUs 165 or one or more RUs 170 can be partially controlled by one or more CUs 160 associated with a donor network entity 105 (e.g., a donor base station 140). One or more donor network entities 105 (e.g., IAB donors) can communicate with one or more additional network entities 105 (e.g., IAB nodes 104) via supported access and backhaul links (e.g., backhaul communication links 120). An IAB node 104 can include an IAB mobile termination (IAB-MT) controlled (e.g., scheduled) by a coupled DU 165 of an IAB donor. The IAB-MT can include a separate set of antennas for relaying communications with UEs 115 or can share the same antennas (e.g., of an RU 170) of the IAB node 104 for accessing via the DU 165 of the IAB node 104 (e.g., referred to as a virtual IAB-MT (vIAB-MT)). In some examples, an IAB node 104 can include a DU 165 that supports a relay chain or configuration (e.g., downstream) of communications links with additional entities (e.g., IAB nodes 104, UEs 115) of an access network. In such cases, one or more components of the disaggregated RAN architecture (e.g., one or more IAB nodes 104 or components of an IAB node 104) can be configured to operate according to the techniques described herein.

[0067] For example, an access network (AN) or RAN can include communications between an access node (e.g., an IAB donor), an IAB node 104, and one or more UEs 115. The IAB donor can facilitate a connection between the core network 130 and the AN (e.g., via a wired or wireless connection to the core network 130). That is, the IAB donor can refer to a RAN node that has a wired or wireless connection to the core network 130. The IAB donor can include a CU 160 and at least one DU 165 (e.g., and RU 170), in which case the CU 160 can communicate with the core network 130 via an interface (e.g., a backhaul link). The IAB donor and the IAB node 104 can communicate via an Fl interface according to a protocol that defines signaling messages (e.g., an Fl AP protocol). Additionally or alternatively, the CU 160 can communicate with the core network via an interface that can be an example of a backhaul link, and can communicate with other CUs 160 (e.g., CUs 160 associated with alternative IAB donors) via an Xn-C interface that can be an example of a backhaul link.

[0068] The IAB node 104 can refer to a RAN node that provides IAB functionality (e.g., for access for UEs 115, wireless self-backhauling capabilities, etc.). The DU 165 can act as a distributed scheduling node toward child nodes associated with the IAB node 104, and the IAB-MT can act as a scheduled node toward parent nodes associated with the IAB node 104. That is, the IAB donor can be referred to as a parent node that communicates with one or more child nodes (e.g., the IAB donor can relay transmissions of a UE through one or more other IAB nodes 104). Additionally or alternatively, the IAB node 104 can also be referred to as a parent node or a child node of other IAB nodes 104 according to a relay chain or configuration of the AN. Thus, the IAB-MT entity of the IAB node 104 can provide a Uu interface for a child IAB node 104 to receive signaling from a parent IAB node 104, and the DU interface (e.g., DU 165) can provide a Uu interface for the parent IAB node 104 to signal to the child IAB node 104 or a UE 115.

[0069] For example, an IAB node 104 can be referred to as a parent node that supports communications for a child IAB node or as a child IAB node that is associated with an IAB donor or both. An IAB donor can include a CU 160 that has a wired or wireless connection (e.g., backhaul communication link 120) to a core network 130 and can act as a parent node for an IAB node 104. For example, a DU 165 of an IAB donor can relay transmissions to a UE 115 through an IAB node 104 or can signal transmissions directly to the UE 115 or both. A CU 160 of an IAB donor can signal communication link establishment to an IAB node 104 via an Fl interface and the IAB node 104 can schedule transmissions (e.g., transmissions relayed from the IAB donor to a UE 115) by a DU 165. That is, data can be relayed to and from an IAB node 104 via signaling via an NR Uu interface to an MT of the IAB node 104. Communications with the IAB node 104 can be scheduled by a DU 165 of an IAB donor and communications with the IAB node 104 can be scheduled by a DU 165 of the IAB node 104.

[0070] In cases where the techniques described herein are applied in the context of a disaggregated RAN architecture, one or more components of the disaggregated RAN architecture can be configured to support a cyclic equivariant inference machine for channel estimation as described herein. For example, some operations described as being performed by a UE 115 or a network entity 105 (e.g., a base station 140) can additionally or alternatively be performed by one or more components of a disaggregated RAN architecture (e.g., an IAB node 104, a DU 165, a CU 160, a RU 170, a RIC 175, a SMO 180).

[0071] A UE 115 can include or can be referred to as a mobile device, a wireless device, a remote device, a handheld device, or a subscriber device, or some other suitable terminology, where the “device” can also be referred to as a unit, a station, a terminal, or a client, among other examples. A UE 115 can also include or can be referred to as a personal electronic device such as a cellular phone, a personal digital assistant (PDA), a tablet computer, a laptop computer, or a personal computer. In some examples, a UE 115 can include or can be referred to as a wireless local loop (WLL) station, an Internet of Things (IoT) device, an Internet of Everything (IoE) device, or a machine type communications (MTC) device, among other examples, which can be implemented in various objects such as appliances or vehicles, among other examples.

[0072] A UE 115 described herein can be able to communicate with various types of devices, such as other UEs 115 that can sometimes act as relays as well as network equipment including base stations 105 and network equipment including macro eNBs or gNBs, small cell eNBs or gNBs, or relay base stations, among other examples, as described herein.Figure 1 As shown.

[0073] The UEs 115 and the network entities 105 can wirelessly communicate with one another using resources associated with one or more carriers using one or more communication links 125 (e.g., access links). The term “carrier” can refer to a set of RF spectrum resources having a defined physical layer structure for supporting communication links 125. For example, a carrier used for a communication link 125 can include a portion of an RF spectrum band (e.g., a bandwidth part (BWP)) operating according to one or more physical layer channels for a given radio access technology (e.g., LTE, LTE-A, LTE-A Pro, NR). Each physical layer channel can carry acquisition signaling (e.g., synchronization signals, system information), control signaling coordinating carrier operation, user data, or other signaling. The wireless communications system 100 can support communication with UEs 115 using carrier aggregation or multi-carrier operation. According to carrier aggregation, a UE 115 can be configured with multiple downlink component carriers and one or more uplink component carriers. Carrier aggregation can be used for both Frequency Division Duplex (FDD) and Time Division Duplex (TDD) component carriers. Communications between a network entity 105 and other devices can refer to communications between these devices and any portion of the network entity 105 (e.g., an entity, sub-entity). For example, the terms “transmit,” “receive,” or “communicate” can refer to any portion of the network entity 105 (e.g., a base station 140, a CU 160, a DU 165, a RU 170) communicating with another device (e.g., directly or via one or more other network entities 105).

[0074] Signal waveforms transmitted via a carrier can be composed of multiple subcarriers (e.g., using multi-carrier modulation (MCM) techniques such as orthogonal frequency division multiplexing (OFDM) or discrete Fourier transform spread OFDM (DFT-S-OFDM)). In a system employing MCM techniques, a resource element can refer to a resource comprising one symbol period (e.g., the duration of one modulation symbol) and one subcarrier, in which case the symbol period and subcarrier spacing can be inversely related. A number of bits carried by each resource element can depend on the modulation scheme (e.g., the order of the modulation scheme, the coding rate of the modulation scheme, or both), such that a relatively large number of resource elements (e.g., in a transmission duration) and a relatively high order modulation scheme can correspond to a relatively high data rate. A wireless communications resource can refer to a combination of a RF spectrum resource, a time resource, and a spatial resource (e.g., spatial layers or beams), and the use of multiple spatial resources can increase the data rate for communications with a UE 115.

[0075] The time interval for network entity 105 or UE 115 can be expressed as a multiple of a basic time unit, such as the sampling period T. s =1 / (Δf) max ·N f ) seconds, where Δf max This can represent the supported subcarrier spacing, while N f This can represent the supported Discrete Fourier Transform (DFT) size. The time interval of the communication resource can be organized according to radio frames, each with a specified duration (e.g., 10 milliseconds (ms)). Each radio frame can be identified by a System Frame Number (SFN) (e.g., ranging from 0 to 1023).

[0076] Each frame may include multiple consecutively numbered subframes or time slots, and each subframe or time slot may have the same duration. In some examples, a frame may (e.g., in the time domain) be divided into subframes, and each subframe may be further divided into a number of time slots. Alternatively, each frame may include a variable number of time slots, and the number of time slots may depend on the subcarrier spacing. Each time slot may include a number of symbol periods (e.g., depending on the length of the cyclic prefix appended to each symbol period). In some wireless communication systems 100, time slots may be further divided into multiple micro-time slots associated with one or more symbols. Excluding the cyclic prefix, each symbol period may be associated with one or more (e.g., N) symbols. f The duration of a symbol period is associated with a ( ) sampling period. The duration of a symbol period can depend on the subcarrier spacing or the operating frequency band.

[0077] A subframe, time slot, micro-time slot, or symbol can be the smallest scheduling unit of the wireless communication system 100 (e.g., in the time domain) and can be referred to as a Transmission Time Interval (TTI). In some examples, the duration of the TTI (e.g., the number of symbol periods in the TTI) can be variable. Additionally or alternatively, the smallest scheduling unit of the wireless communication system 100 can be dynamically selected (e.g., in a burst of shortened TTIs (sTTIs)).

[0078] Physical channels can be multiplexed according to various techniques to use a carrier for communication. For example, one or more of time division multiplexing (TDM) techniques, frequency division multiplexing (FDM) techniques, or hybrid TDM-FDM techniques can be used to multiplex physical control channels and physical data channels for transmission via a downlink carrier. A control region (e.g., a control resource set (CORESET)) for a physical control channel can be defined by a collection of symbol periods having the same or different durations, and can extend across the entire bandwidth of a carrier or a subset of the

[0079] In some examples, network entities 105 (e.g., base stations 140, RUs 170) can be mobile and thus provide communication coverage for a moving coverage area 110. In some examples, different coverage areas 110 associated with different technologies can overlap, but different coverage areas 110 can be supported by the same network entity 105. In some other examples, overlapping coverage areas 110 associated with different technologies can be supported by different network entities 105. Wireless communications system 100 can include, for example, a heterogeneous network in which different types of network entities 105 provide coverage for various coverage areas 110 using the same or different radio access technologies.

[0080] Wireless communications system 100 can be configured to support ultra-reliable communications or low-latency communications, or various combinations thereof. For example, wireless communications system 100 can be configured to support ultra-reliable low-latency communications (URLLC). UEs 115 can be designed to support ultra-reliable or low-latency or critical functions. Ultra-reliable communications can include private communication or group communication, and can be supported by one or more services such as push-to-talk, video, or data. Support for ultra-reliable, low-latency functions can include prioritization of services, and such services can be used for public safety or general commercial applications. The terms ultra-reliable, low-latency, and ultra-reliable low-latency can be used interchangeably herein.

[0081] In some examples, UEs 115 can be configured to communicate directly with other UEs 115 via device-to-device (D2D) communication links 135 (e.g., according to a peer-to-peer (P2P), D2D, or sidelink protocol). In some examples, one or more UEs 115 in a group that is performing D2D communication can be within the coverage area 110 of a network entity 105 (e.g., base station 140, RU 170) that supports aspects of such D2D communication configured by the network entity 105 (e.g., scheduled). In some examples, one or more UEs 115 in such a group can be outside the coverage area 110 of a network entity 105 or can otherwise be unable to receive transmissions from the network entity 105. In some examples, the group of UEs 115 communicating via D2D communication can support a one-to-many (1 :M) system, where each UE 115 transmits to every other UE 115 in the group. In some examples, a network entity 105 can facilitate scheduling of resources for D2D communication. In some other examples, D2D communication can be carried out between UEs 115 without involvement of a network entity 105.

[0082] The core network 130 can provide user authentication, access authorization, tracking, Internet Protocol (IP) connectivity, and other access, routing, or mobility functions. The core network 130 can be an evolved packet core (EPC) or 5G core (5GC), which can include at least one control plane entity that can manage access and mobility (e.g., a mobility management entity (MME), an access and mobility management function (AMF)) and at least one user plane entity that can route packets or

[0083] Wireless communications system 100 can operate using one or more frequency bands, often in the range of 300 megahertz (MHz) to 300 gigahertz (GHz). Generally, the region from 300 MHz to 3 GHz is known as the ultra-high frequency (UHF) region or decimeter band, since the wavelengths range from approximately one decimeter to one meter in length. The UHF wave s can be blocked or redirected by buildings and environmental features, but the waves can penetrate structures sufficiently for a macro cell to provide service within an indoor location. The use of UHF frequencies, however, may

[0084] Wireless communications system 100 can utilize both licensed and unlicensed RF spectrum bands. For example, wireless communications system 100 can employ License Assisted Access (LAA), LTE-Unlicensed (LTE-U) radio access technology, or NR technology in an unlicensed

[0085] The electromagnetic spectrum is often subdivided based on frequency / wavelength into various classes, bands, channels, and so forth. In 5G NR, two initial operating bands have been identified as frequency range designations FR1 (410 MHz - 7. 125 GHz) and FR2 (24.25 GHz - 52.6 GHz). It should be understood that although a portion of FR1 is greater than 6 GHz, FR1 is often referred to (interchangeably) as a “Sub-6 GHz” band in various documents and articles. A similar nomenclature issue sometimes occurs with regard to FR2, which is often referred to (interchangeably) as a “millimeter wave” band in documents and articles, despite being different from the extremely high frequency (EHF) band (30 GHz - 300 GHz) which is identified by the International Telecommunications Union (ITU) as a “millimeter wave” band.

[0086] The frequencies between FR1 and FR2 are often referred to as mid-band frequencies. Recent 5G NR studies have identified operating bands for these mid-band frequencies as Frequency Range designation FR3 (7.125 GHz - 24.25 GHz). Bands that fall within FR3 can inherit FR1 and / or FR2 characteristics, and thus can effectively extend the features of FR1 and / or FR2 to mid-band frequencies. Moreover, higher bands are currently being explored to extend 5G NR operations beyond 52.6 GHz. For example, three higher operating bands have been identified as Frequency Range designations FR4-a or FR4-1 (52.6 GHz - 71 GHz), FR4 (52.6 GHz - 114.25 GHz), and FR5 (114.25 GHz - 300 GHz). Each of these higher bands fall within the EHF band.

[0087] In light of the above aspects, unless specifically stated otherwise, it should be appreciated that the term “sub-6 GHz” or the like, if used herein, can broadly represent frequencies that can be less than 6 GHz, can be within FR1, or can include mid-band frequencies. Further, unless specifically stated otherwise, it should be appreciated that the term “millimeter wave” or the like, if used herein, can broadly represent frequencies that can include mid-band frequencies, can be within FR2, FR4, FR4-a or FR4-1, and / or FR5, or can be within the EHF band.

[0088] The network entities 105 (e.g., base stations 140, RUs 170) or UEs 115 can be equipped with multiple antennas, which can be used to employ techniques such as transmit diversity, receive diversity, MIMO communication, or beamforming. The antennas of a network entity 105 or a UE 115 can be located in one or more antenna arrays or antenna panels, which can support MIMO operations or transmit or receive beamforming. For example, one or more base station antennas or antenna arrays can be co-located at an antenna assembly, such as an antenna tower. In some examples, the antennas or antenna arrays associated with a network entity 105 can be located at different geographic locations. A network entity 105 can include an antenna array with a set of multiple rows and multiple columns of antenna ports that the network entity 105 can use for beamforming to support communication with UEs 115. Likewise, a UE 115 can include one or more antenna arrays, which can support various MIMO or beamforming operations. Additionally or alternatively, an antenna panel can support RF beamforming for signals transmitted via the antenna ports.

[0089] The network entity 105 or the UE 115 can use MIMO communications to exploit multipath signal propagation to increase the spectral efficiency and / or combat the effects of fading. Such techniques can be referred to as spatial multiplexing. The multiple signals may, for example, be transmitted by a transmitting device via different antennas or different combinations of antennas. Similarly, the multiple signals can be received by a receiving device via different antennas or different combinations of antennas. Each of the multiple signals can be referred to as a separate spatial stream, and can carry

[0090] Beamforming, which can also be referred to as spatial filtering, directional transmission, or directional reception, is a signal processing technique that can be used at a transmitting device or a receiving device (e.g., a network entity 105, a UE 115) to shape or steer a beam of energy in a specific direction, for example, along with the spatial path between the transmitting device and the receiving device. Beamforming can be achieved by combining the signals communicated by antennas of an antenna array such that some signals propagating at different directions experience constructive interference while others experience destructive interference. The adjustment of one or more signals by the transmitting device or receiving device can include amplifying the signal, changing a phase of the signal, or changing a frequency of the signal. The adjustments can be made by the components of the transmitting device or the receiving device that are used to perform the

[0091] The network entity 105 or UE 115 can use beam sweeping techniques as part of a beamforming operation. For example, a network entity 105 (e.g., a base station 140, a RU 170) can use multiple antennas or antenna arrays (e.g., antenna panels) to conduct a beamforming operation for directional communications with a UE 115. Some signals (e.g., synchronization signals, reference signals, beam- selection signals, or other control signals) can be transmitted by a network entity 105 multiple times in different directions. For example, a network entity 105 can transmit a signal according to different beamforming weight sets associated with different directions of transmission. The transmissions in the different beam directions can be used by a transmitting device (such as a network entity 105) or by a receiving device (such as a UE 115) to identify a beam direction for subsequent transmission or reception by the network entity 105.

[0092] Some signals, such as data signals associated with a particular receiving device, can be transmitted by a transmitting device (e.g., a transmitting network entity 105, a transmitting UE 115) in a single beam direction (e.g., a direction associated with the receiving device, such as a receiving network entity 105 or a receiving UE 115). In some examples, the beam direction associated with transmissions in a single beam direction can be determined based on a signal that was transmitted in one or more beam directions. For example, a UE 115 can receive one or more of the signals transmitted by the network entity 105 in different directions, and can report to the network entity 105 an indication of the signal that the UE 115 received with a highest signal quality, or other acceptable signal quality.

[0093] In some examples, transmissions by a device (e.g., by a network entity 105 or a UE 115) can be performed using multiple beam directions, and the device can use a combination of digital precoding or beamforming to generate a combined beam for transmissions (e.g., from a network entity 105 to a UE 115). A UE 115 can report feedback that indicates precoding weights for one or more beam directions, and the feedback can correspond to a configured set of beams across a system bandwidth or one or more sub-bands. The network entity 105 can transmit a reference signal (e.g., a cell-specific reference signal (CRS), a channel state information reference signal (CSI-RS)), which can or can not be precoded. The UE 115 can provide feedback for beam selection, which can be a precoding matrix indicator (PMI) or codebook-based feedback (e.g., a multi-panel type codebook, a linear combination type codebook, a port selection type codebook). Although the techniques are described with reference to signals transmitted by a network entity 105 (e.g., base stations 140, RUs 170) in one or more directions, a UE 115 can use similar techniques for transmitting signals multiple times in different directions (e.g., for identifying a beam direction for subsequent transmission or reception by the UE 115), or for transmitting a signal in a single direction (e.g., for transmitting data to a receiving device).

[0094] A receiving device (e.g., a UE 115) can perform reception operations according to a number of receive configurations (e.g., directional listening) when receiving various signals from a transmitting device (e.g., a network entity 105), such as synchronization signals, reference signals, beam selection signals, or other control signals. For example, a receiving device can perform reception according to a number of receive directions with a set of antennas, different

[0095] Wireless communications system 100 can be a packet-based network that operates according to a layered protocol stack. In the user plane, communications at the bearer or PDCP layer can be IP -based. A RLC layer can perform packet segmentation and reassembly to communicate over logical channels. The MAC layer can perform priority handling and multiplexing of logical channels into transport channels. The MAC layer can also implement error detection techniques, error correction techniques, or both to support retransmissions to improve link efficiency. In the control plane, the RRC layer can provide establishment, configuration, and maintenance of an RRC connection between a UE 115 and a network entity 105 or core network 130 supporting radio bearers for user plane data. The PHY layer can map transmission channels to physical channels.

[0096] In some cases, wireless communications system 100 can support SISO communications, MIMO communications, or both. For example, SISO communications can include communications between a single transmitter and a single receiver, while MIMO communications can include communications between multiple transmitters and multiple receivers. In some cases, MIMO communications can include multiple SISO communications via transmitter and receiver pairs (e.g., pairs of transmit and receive antennas). For example, a network entity 105 can include a first transmit antenna and a second transmit antenna, and a UE 115 can include a first receive antenna and a second receive antenna. The pairs of antennas can include a first transmission utilizing the first receive antenna, a first transmission utilizing the second receive antenna, a second transmission utilizing the first receive antenna, and a second transmission utilizing the second receive antenna. Each pair of antennas can modify a signal prior to signal transmission with a beamforming matrix (e.g., a precoding matrix, an orthogonal matrix) to minimize interference (e.g., a linear precoder, a beamformer can create a beam that focuses energy for each receive antenna by weighting the phase and magnitude of the transmit antennas). Because the beamforming matrix can be unknown to the receiver (e.g., UE 115, network entity 105), the receiver can estimate the precoding channel.

[0097] In some examples, wireless communications system 100 can support channel estimation. For example, channel estimation can be used for resource grid (e.g., slot-based) based wireless MIMO-OFDM systems. Channel estimation can be 5G NR channel estimation with varying DMRS patterns, number of resource blocks, etc. Channel estimation can be used for sparse observation based super-resolution or signal recovery.

[0098] In MIMO communications, multiple layers of information (e.g., communications) can interfere with each other, making channel estimation more complex. In some cases, orthogonal cover codes can be used to remove the interference through a despreading step. However, the despreading step can be insufficient for frequency selective or fast fading channels (e.g., high delay spread and high Doppler response). Additionally, narrowband MIMO communications (e.g., communications using blocks of a relatively small bandwidth part) can use different precoding matrices for each resource group (e.g., physical resource groups (PRGs)) that can be unknown to a UE 115. Unknown precoding matrices can add higher complexity to channel estimation, such that some wireless communication systems can not use the correlation between non-contiguous PRGs in channel estimation. For example, some estimation techniques (e.g., MMSE) can utilize TRS or another contiguous reference signal to compute channel properties (e.g., Doppler, delay spread, SNR, etc.) and estimate the channel resources based on the channel properties.

[0099] In some cases, some estimation techniques can include least squares and linear MMSE (LMMSE). Least squares can not use information about channel statistics or noise variance and can not model the correlation across different PRGs and MIMO layers, thus making relatively simple to implement with low computational overhead. However, estimation accuracy can be insufficient for most use cases (e.g., practical applications). LMMSE can utilize second order channel statistics and noise variance (e.g., bin-based strategies based on estimated channel parameters such as Doppler, delay spread, etc.). In some conditions, LMMSE can have high computational expense and have relatively low estimation error. However, LMMSE can not model the correlation across different PRGs and MIMO layers.

[0100] In some cases, some estimation techniques can include deep learning based techniques. Deep learning based techniques can not utilize explicit information of channel statistics and can utilize non-linear interpolation. Unlike LMMSE, deep learning techniques can not utilize large matrix inversion operations. Deep learning techniques can utilize separate networks for each DMRS pattern. In some cases, deep learning techniques can not consider the correlation across different PRGs.

[0101] The techniques described herein allow for the use of a cyclic variational inference machine to compute channel estimates, which can result in channel estimates based on DMRS symbols that exploit the correlation between non-contiguous PRGs (e.g., agnostic to precoder). In some cases, a wireless communication system 100 (e.g., an OFDM system) can deploy pilot-based channel estimation techniques to obtain CSI relatively accurately (e.g., obtaining accurate CSI can help maintain high data throughput, for example, in fast fading environments). The pilot signals can be referred to as DMRS symbols. DMRS symbols can be inserted into a transmission time slot according to a known DMRS pattern, allowing a wireless device to estimate the unknown resources of a channel (e.g., non-DMRS locations, resources allocated for data signals) based on known resources (e.g., DMRS symbols). In some cases, the DMRS pattern can be preconfigured (e.g., a fixed set of possible DMRS patterns). The DMRS pattern can be used based on channel characteristics.

[0102] In some examples, a wireless device can receive an assignment of a set of resources associated with a channel, where the set of resources includes a first subset of resources allocated for data transmission (e.g., non-DMRS) and a second subset of resources allocated for reference signals (e.g., DMRS). The wireless device can generate a plurality of channel estimates (e.g., SISO channel estimates) for each layer of the channel and perform a refinement operation utilizing the estimates to generate a channel estimate associated with the multiple layers (e.g., a MIMO channel estimate). In some cases, the refinement operation can include a plurality of iterations. For example, each iteration can include generating a respective gradient associated with each of the per-layer channel estimates based on the per-layer channel estimates and observed resources of the channel (e.g., known DMRS resources); generating a current set of values of latent variables (e.g., inference variables based on observed variables) based on a previous set of values of the latent variables, the respective gradient, and the per-layer channel estimates; and modifying (e.g., refining, updating, improving) the channel estimate based on the current set of values of the latent variables, the per-layer channel estimates, and the respective gradient. In some cases, the refinement operation can be performed by a refinement network that includes a likelihood module, an encoder module, and a decoder module.

[0103] In some cases, the wireless communication system 100 can incorporate an end-to-end (E2E) use of neural networks for channel state feedback. Such neural network structures can be used for channel state information feedback (CSF) by providing an intermediate channel representation to a wireless communication network, and a wireless device (e.g., a receiving wireless device, a network entity 105) can reconstruct the channel (e.g., the proposed method is a separate implementation at the UE and network side), such that this type of model architecture can have some degree of specification for interoperability.

[0104] Figure 2The network architecture 300 (e.g., disaggregated base station architecture, disaggregated RAN architecture) that supports cyclic equivariant reasoning machine for channel estimation in accordance with one or more aspects of the present disclosure is illustrated. The network architecture 200 can illustrate an example for implementing one or more aspects of the wireless communication system 100. The network architecture 200 can include one or more CUs 160-a that can communicate directly with the core network 130-a via a backhaul communication link 120-a or indirectly with the core network 130-a through one or more disaggregated network entities 105 (e.g., near-RT RIC 175-b via an E2 link or non-RT RIC 175-a associated with an SMO 180-a (e.g., SMO framework), or both). The CU 160-a can communicate with one or more DUs 165-a via respective fronthaul communication links 162-a (e.g., Fl interface). The DU 165-a can communicate with one or more RUs 170-a via respective front-haul communication links 168-a. The RU 170-a can be associated with a respective coverage area 110-a and can communicate with a UE 115-a via one or more communication links 125-a. In some implementations, the UE 115-a can be served by multiple RUs 170-a simultaneously.

[0105] Each of the network entities 105 (e.g., CU 160-a, DU 165-a, RU 170-a, non-RT RIC 175-a, near-RT RIC 175-b, SMO 180-a, Open Cloud (O-Cloud) 205, Open eNB (O-eNB) 210) of the network architecture 200 can include or be coupled with one or more interfaces configured to receive or transmit signals (e.g., data, information) via a wired or wireless transmission medium. An associated processor (e.g., controller) of each network entity 105 or interface providing instructions to the network entity 105 can be configured to communicate with one or more of the other network entities 105 via the transmission medium. For example, the network entities 105 can include a wired interface configured to receive signals on or transmit signals on a wired transmission medium to one or more of the other network entities 105. Additionally or alternatively, the network entities 105 can include a wireless interface that can include a receiver, a transmitter, or a transceiver (e.g., an RF transceiver) configured to receive signals on or transmit signals on a wireless transmission medium to one or more of the other network entities 105, or both.

[0106] In some examples, the CU 160-a can host one or more higher layer control functions. Such control functions can include RRC, PDCP, SDAP, etc. Each control function can be implemented with an interface configured to communicate signals with other control functions hosted by the CU 160-a. The CU 160-a can be configured to handle user plane functions (e.g., CU-UP), control plane functions (e.g., CU-CP), or a combination thereof. In some examples, the CU 160-a can be logically split into one or more CU-UP units and one or more CU-CP units. When implemented in an O-RAN configuration, the CU-UP units can communicate bi-directionally with the CU-CP units via an interface, such as an El interface. The CU 160-a can be implemented to communicate with the DU 165-a as needed for network control and signaling.

[0107] The DU 165-a can correspond to a logical unit that includes one or more functions (e.g., base station functions, RAN functions) for controlling the operation of one or more RUs 170-a. In some examples, the DU 165-a can host, at least in part, one or more of the RLC layer, the MAC layer, and the PHY layer, or one or more aspects thereof (e.g., high PHY layers, such as modules for FEC encoding and decoding, scrambling, modulation and demodulation, etc.), depending at least in part on a functional split, such as those defined by the Third Generation Partnership Project (3 GPP). In some examples, the DU 165-a can also host one or more low PHY layers. Each layer can be implemented with an interface configured to communicate signals with other layers hosted by the DU 165-a or with control functions hosted by the CU 160-a.

[0108] In some examples, lower layer functions can be implemented by one or more RUs 170-a. For example, an RU 170-a controlled by the DU 165-a can correspond to a logical node that hosts RF processing functions or low PHY layer functions (e.g., performing fast Fourier transform (FFT), inverse FFT (iFFT), digital beamforming, physical random access channel (PRACH) extraction and filtering, etc.), or both, based at least in part on a functional split, such as a lower layer functional split. In such an architecture, the RU 170-a can be implemented to handle over-the-air (OTA) communications with one or more UEs 115-a. In some implementations, real-time and non-real-time aspects of control plane and user plane communications with the RU 170-a can be controlled by the corresponding DU 165-a. In some examples, such a configuration can enable the DU 165-a and the CU 160-a to be implemented in a cloud-based RAN architecture, such as a vRAN architecture.

[0109] The SMO 180-a can be configured to support RAN deployment and provisioning of non-virtualized and virtualized network entities 105. For non-virtualized network entities 105, the SMO 180-a can be configured to support deployment of dedicated physical resources for RAN coverage requirements, which can be managed via an operations and maintenance interface (e.g., an Ol interface). For virtualized network entities 105, the SMO 180-a can be configured to interact with a cloud computing platform (e.g., O-Cloud 205) via a cloud computing platform interface (e.g., an 02 interface) to perform network entity lifecycle management (e.g., to instantiate a virtualized network entity 105). Such virtualized network entities 105 can include, but are not limited to, CUs 160-a, DUs 165-a, RUs 170-a, and near-RT RICs 175-b. In some implementations, the SMO 180-a can communicate with components configured according to 4G RAN (e.g., via an Ol interface). Additionally, or alternatively, in some implementations, the SMO 180-a can communicate directly with one or more RUs 170-a via an Ol interface. The SMO 180-a can also include a non-RT RIC 175-a configured to support the functionality of the SMO 180-a.

[0110] The non-RT RIC 175-a can be configured to include logical functions that enable non-real-time control and optimization of RAN elements and resources, artificial intelligence (Al) or machine learning (ML) workflows, including model training and update, or policy-based steering of applications / features in the near-RT RIC 175-b. The non-RT RIC 175-a can be coupled to, or in communication with, the near-RT RIC 175-b (e.g., via an Al interface). The near-RT RIC 175-b can be configured to include logical functions that enable near-real-time control and optimization of RAN elements and resources via data collection and actions on interfaces that connect one or more CUs 160-a, one or more DUs 165-a, or both, and the near-RT RIC 175-b (e.g., via an E2 interface).

[0111] In some examples, to generate an AI / ML model to be deployed in near-RT RIC 175-b, non-RT RIC 175-a can receive parameters or external enrichment information from an external server. Such information can be utilized by near-RT RIC 175-b and can be received at SMO 180-a or non-RT RIC 175-a from non-network data sources or from network functions. In some examples, non-RT RIC 175-a or near-RT RIC 175-b can be configured to tune RAN behavior or performance. For example, non-RT RIC 175-a can monitor long-term trends and patterns of performance and employ an AI model or ML model to perform corrective actions by SMO 180-a (e.g., via reconfiguration of Ol) or via generation of RAN management policies such as Al policies.

[0112] Figure 3 An example of a network 300 that supports cyclic equivariant inference machine for channel estimation is shown, in accordance with one or more aspects of the present disclosure. In some examples, network 300 can be implemented by aspects of wireless communication system 100. For example, network 300 can be implemented by a UE 115, a network entity 105, or both, as described herein with reference to FIGs. 1-2. Figure 1 and Figure 2

[0113] In some examples, a wireless device (e.g., UE 115, network entity 105) can transmit a signal via one or more slots (e.g., frequency and time grid) using a plurality of resources (e.g., time resources, frequency resources, etc.). A resource element (e.g., symbol 315, DMRS 320, a single subcarrier for a single OFDM symbol) can be grouped with a plurality of resource elements to form a physical resource block (PRB) (e.g., PRB 310). Each column of resources of a PRB 310 (e.g., resource elements of the same time resource and different frequency resources) can be considered a single resource block (e.g., OFDM symbol). In some cases, a PRB 310 can include twelve subcarrier frequencies (of the frequency domain) and fourteen OFDM symbols (of the time domain). A plurality of PRBs 310 can be bundled (e.g., grouped, combined) to form a single PRG, such as PRG 305. In some cases, a PRG 305 can include a number of PRBs 310 determined by a bundle size parameter (e.g., bundleSize, a number of contiguous PRBs stacked together in a single PRG). For example, a bundle size parameter can indicate two PRBs 310 or four PRBs 310 (e.g., four contiguous resource blocks) for a narrowband precoding operation or zero PRBs 310 (e.g., no stacked PRBs) for a wideband precoding operation, and form a portion of a bandwidth part.

[0114] ​In some cases, a wireless communications system (e.g., the wireless communications system 100) can utilize pilot-based channel estimation techniques. Pilot symbols can be referred to as DMRS symbols (e.g., the DMRS symbols 320). For example, a wireless device can transmit a signal including one or more PRBs 310 via a physical downlink shared channel (PDSCH) (e.g., a channel for user data). The PRBs 310 can include various symbols 315 (e.g., resource elements allocated for a data signal), various DMRS symbols 320 (e.g., resource elements configured for a reference signal for demodulation), and, in some cases, null symbols (e.g., resource elements without data allocated).

[0115] The DMRS symbols 320 can be inserted in various resource elements of the PRBs 310 according to a resource configuration pattern (e.g., a DMRS pattern). For example, a wireless device can be configured with various DMRS patterns and select a DMRS pattern 345 for the signal. In some cases, the various DMRS patterns can include DMRS symbols inserted in adjacent resource elements, every other resource element, a single resource block of a PRB, multiple resource blocks of a PRB, and other potential configurations.

[0116] In some examples, the DMRS pattern 345 can be known by a receiving wireless device. For example, a receiving wireless device can receive the signal and determine which symbols of the PRGs 305 include various DMRS 320 based on the DMRS pattern 345. To extract (e.g., process, determine, decode, estimate) data at the symbols 315, the receiving wireless device can perform a channel estimation process. For example, according to Equation 1, a received DMRS 320 (e.g., y i ) can be equal to a combination between noise (e.g., interference, n i ) and a product between a channel (e.g., PDSCH, h i ) and original data (e.g., x i ).

[0117] Equation 1: y i = h i ⊙x i + n i

[0118] Because the DMRS 320 is a pilot symbol known to both a transmitting wireless device (e.g., a UE 115, a network entity 105) and a receiving wireless device (e.g., a UE 115, a network entity 105), the receiving wireless device can estimate h i given some noise) based on the known y i and x i(e.g., extract the channel at the location of the DMRS). The receiving wireless device can then interpolate (e.g., inpaint for an image) the estimated channel across the various symbols 315 (e.g., the remaining resource elements) to extract (e.g., compute) the data at the symbols 315. In some cases, n i The signal can include cross-MIMO interference, inter-PRG interference, intra-PRG interference, etc. Some channel estimation techniques can not account for (e.g., compute) these types of interference (e.g., noise), which can result in inaccurate channel estimates and inaccurate data estimates.

[0119] In some implementations, the signal can include multiple PRGs 305, where each PRG 305 can be configured (e.g., precoded) according to a unique precoding matrix For example, the signal can include four PRGs 305, each precoded according to a unique precoding matrix. The unique precoding matrix can be found by performing singular value decomposition (SVD) on the resource blocks within the PRG 305 or as a random (e.g., orthogonal) precoder. In some cases, the effective channel (e.g., ) of the first PRG 305 can be equal to the product between the channel and the unique precoding matrix (v prg ) of the first PRG 305.

[0120] Equation 2: In some cases, the receiving wireless device can not know which precoding matrix is applied to which PRG 305, and thus, the unique precoding of the channel for each PRG 305 can prohibit smooth interpolation of the channel between PRGs 305.

[0121] In some implementations, MIMO communications can introduce additional complexity to the channel estimation formula (e.g., SISO channel estimation formula). For example, at the DMRS tone inference (e.g., extraction) step, the formula can include multiple unknown equations (e.g., multiple equations with multiple unknown variables), which can result in solving an ill-posed inverse problem. The complexity can include determining the alignment of the channel across different PRG beams and then exploiting the correlation across them. Additionally, the multiple layers can add additional multiplexing information (e.g., correlation across receiver and transmitter antenna pairs) that can be used to enhance channel estimation performance.

[0122] The techniques described herein allow for the use of a cyclic equivariant reasoner to compute channel estimates. For example, a single neural network estimator is used for multiple use cases of channel estimation including modular and interpretable model design (e.g., channel profile estimation, various DMRS pattern configurations, SNR, cross-MIMO estimation, inter-PRG estimation, intra-PRG estimation, etc.). In some implementations, channel estimation can include multiple stages (e.g., steps). For example, a first stage can include solving for SISO channel estimates (e.g., for each transmitter and receiver antenna pair). A second stage can include using the SISO channel estimates to solve for MIMO channel estimates (e.g., learning correlations between antenna pairs). A third stage can include solving for MIMO channel estimates within a PRG beam (e.g., intra-PRG, learning correlations between resource elements within each PRG). A fourth stage can include solving for MIMO channel estimates across PRG beams (e.g., inter-PRG, learning correlations between resource elements across PRGs).

[0123] Additionally or alternatively, in some examples as described herein, channel estimates can be computed using a cyclic equivariant reasoner and based on MMSE operations. MMSE operations can be an estimation technique that utilizes linear equalization to estimate a channel. MMSE can be supported by hardware within a wireless device, which can enable reduced complexity and processing as compared to performing initial estimates based on machine learning. A cyclic equivariant reasoner can represent an example of a type of machine learning model that provides relatively reliable and accurate estimates based on a given set of inputs. For example, a first stage of channel estimation can include solving for SISO channel estimates (e.g., for each transmitter and receiver antenna pair) based on MMSE operations 350 (e.g., which can also be referred to as average MMSE (AMMSE) operations). A first set of multiple channel estimates for each layer of a channel can be generated based on the MMSE operations 350. In this case, the remaining segments can build on the MMSE generated estimates. In some cases, the various stages of channel estimation can be performed iteratively (e.g., including multiple iterations of the various stages). In some cases, one or more of the stages can utilize SNR estimates (e.g., genie values). Although four stages are described, a channel estimation process utilizing the described techniques can include more or fewer stages, stages including various other steps, stages without one or more of the described steps, or any combination thereof. While the stages are described as four separate stages, they can be considered as one continuous process.

[0124] In some cases, a network can perform the various stages. For example, the network can be a coarse network as described herein with reference to Figure 3 and a fine network as described herein with reference to Figures 4 to 6The refined network. In some cases, the network can include u-net type (e.g., u-net 525) encoder (e.g., encoder 430) and decoder (e.g., decoder 435) convolutional blocks followed by an attention-based (e.g., attention 605) refined network for longer range correlation. In some examples, the coarse network 325 can provide an initial channel estimate (e.g., coarse estimate) per PRG per antenna pair (e.g., transmitter and receiver antenna pair) by learned interpolation (e.g., smoothing). Additionally or alternatively, the MMSE operation 350 can provide an initial channel estimate per PRG per antenna pair, and the coarse network 325 can modify or update the channel estimate by learned interpolation.

[0125] The network 300 can represent a first stage. A wireless device (e.g., as described herein with reference to a UE 115, a network entity 105, or both) can receive an assignment of a set of resources associated with a channel (e.g., a PDSCH channel). The set of resources can include a first subset of resources (e.g., symbols 315) allocated for a data signal and a second subset of resources (e.g., DMRS symbols 320) allocated for a reference signal. Figure 1

[0126] In some cases, the wireless device can perform the MMSE operation 350 to generate a first set of multiple channel estimates 355 (e.g., LS channel estimates at the DMRS symbols 320) associated with respective layers 360 of the channel for the set of resources. For example, the set of resources can include one or more PRGs 305, and the second subset of resources can include individual DMRS symbols 320 inserted into respective resource elements based on one resource configuration pattern (e.g., pattern 345) of a set of resource configuration patterns (e.g., a set of DMRS patterns). In some examples, one or more resources of the first set of resources including non-DMRS symbols 315 can be initialized with zero entries. Each respective layer 360 can be associated with a respective antenna pair (e.g., transmitter and receiver pair) of a plurality of SISO antenna pairs. In some implementations, initial values of the first set of multiple channel estimates 355 can be associated with the SISO antenna pairs. The MMSE operation 350 can be associated with relatively low computational cost and processing compared to other channel estimation techniques. As described herein, the network 300 can utilize the MMSE channel estimates as input and can build upon the MMSE estimates. For example, the MMSE operation 350 can be followed by one or more attention-based refinement operations (e.g., machine learning-based channel estimation) for relatively long range correlation, which can reduce complexity and processing compared to other channel estimation techniques.

[0127] ​After performing the MMSE operation 350, the wireless device can use the first set of channel estimates 355 to generate a second set of channel estimates 335 using a network 325 (e.g., a coarse network). In some examples, the network 325 can include or be based on a machine learning model for channel estimation. In some examples, the network 325 can perform a non-linear two-dimensional interpolation of the channel based on the first set of channel estimates 335. For example, the network 325 can interpolate the channel in time and frequency domain (e.g., two dimensions).

[0128] The network 325 can involve (e.g., input) the first set of channel estimates 355 generated by the MMSE operation 350 based on the PRGs 305. The network 325 can perform various iterations 330 on the first set of channel estimates 355. For example, the network 325 can include a u-net encoder-decoder fully convolutional network, where each iteration can include gated and gated dilated convolutional units. In some cases, for one or more iterations 330, the network 325 can replicate and concatenate the results of a previous iteration 330 to the one or more iterations 330. In some examples, the first set of channel estimates 355 and the second set of channel estimates 335 can be channel estimates for respective SISO antenna pairs per PRG beam (e.g., where h is the channel estimate, m is the PRG index, k is the PRB index, i,j is the MIMO index for a given antenna pair and (f,t) is the resource element index within the PRB two-dimensional grid). For example, the wireless device can receive a respective PRG beam per antenna pair. If the MIMO communication includes two transmit antennas and two receive antennas, there can be four antenna pairs, each with a number of PRG beams. The wireless device can utilize the MMSE operation 350 to generate a first channel estimate 355 for each antenna pair and each PRG beam 305, and utilize the network 325 to further generate a second channel estimate 335 for each antenna pair and each PRG beam 305. In some cases, the network 325 can output The network 325 can additionally generate latent variables. For example, the latent variables can be estimates 340 (e.g., z estimates) that can not be directly observed, but are inferred from other observed parameters. The latent variables can be an abstract representation of underlying channel characteristics (e.g., Doppler shift, delay spread). The network 325 (e.g., an embedding network) can serve as a feature extractor to produce initial latent values (e.g., z τ＝0 ), which can be further used by subsequent refinement modules. In some examples, the output of the network 325 can be an input (e.g., input 415) to the network 400.

[0129] In some cases, the techniques described herein can result in various advantages over other channel estimation techniques. For example, channel estimation by a cyclic equivariant reasoning machine (e.g., networks 300, 400, 500, and 600) can provide various signal processing and deep learning advantages. For example, signal processing can be based on DMRS (e.g., excluding dependence on TRS, except for SNR), excluding explicit parameter estimation (e.g., Doppler shift, delay spread, etc.), avoiding legacy binning strategies, leveraging relatively less memory and computational overhead (e.g., reducing maintenance of parameter banks), modeling additional interactions (e.g., interference, cross-MIMO, intra-PRG, inter-PRG processing gains), abstracting orthogonal cover code (OCC) despreading steps (as described herein with reference to Figure 1 The described), or any combination thereof, thereby circumventing associated computational cost and performance loss. Deep learning techniques can include a variable number of PRG beams, a variable number of beam sizes (e.g., PRBs per PRG), multiple DMRS patterns (e.g., multiple input DMRS configurations, number of additional columns, configuration type), underlying mathematical symmetries, forward models into network design, and modular and interpretable architecture (e.g., with potential to perform ablation studies and measure component significance). By leveraging network 325 building on MMSE operation 350 as described herein, one or more non-linear interpolation gains can be added to the MMSE estimate. MMSE operation 350 can provide a partial machine learning channel estimation solution that enables machine learning capabilities while maintaining existing channel estimation and demapping hardware at a wireless device, which can reduce memory consumption and computational cost, among other possibilities.

[0130] Figure 4 An example of a network 400 that supports cyclic equivariant reasoning machine for channel estimation in accordance with one or more aspects of the present disclosure is shown. In some examples, network 400 can be implemented by aspects of wireless communication system 100. For example, network 400 can be implemented by a UE 115, a network entity 105, or both, as described herein with reference to Figure 1 and Figure 2 .

[0131] In some cases, network 400 can support channel estimation. For example, a channel estimation problem can be defined as maximizing the posterior of a channel (e.g., H) given an observed signal (e.g., y) and a signal (e.g., x), as described herein with reference to In some cases, a conditional probability distribution (e.g., P(H|y,x)) can be parameterized by channel characteristics (e.g., delay-Doppler profile). As described herein with reference to Figure 3 , a latent variable (e.g., z) can be an abstract representation of underlying channel characteristics (e.g., Doppler, delay spread, etc.).

[0132] In some cases, the output of network 300 can be the input of network 400. For example, input 415 can include one or more channel estimates (e.g., a channel estimate 335 per SISO antenna pair per PRG 305) and corresponding estimates 340 (e.g., a corresponding latent variable per channel estimate). Input 415 can be an input to refinement network 405, which can include various iterations 410. For example, a wireless device (e.g., network entity 105, UE 115) can perform a refinement operation on the channel estimates (e.g., via refinement network 405). The refinement operation can include multiple iterations of generating a corresponding gradient (e.g., gradient 480) based on the channel estimates and observations of measurements of the resource subset (e.g., DMRS symbols 320) to the resource subset (e.g., DMRS symbols), as described herein with reference to Figure 4 the (e.g., module 425); generating a second set of latent variables based on the channel estimates and the corresponding gradients, as described herein with reference to Figure 4 and Figure 5 the (e.g., module 430); and modifying the channel estimates associated with the plurality of layers based on the second set of latent variables, the channel estimates, and the corresponding gradients, as described herein with reference to Figure 5 the (e.g., module 435).

[0133] In some cases, refinement network 405 can include iterative refinements by various refinement units. For example, each iteration 410 of refinement network 405 can be performed by various modules (e.g., three or four unique refinement modules). For example, iteration 410 can include module 425 (e.g., a likelihood module), module 430 (e.g., an encoder module), and module 435 (e.g., a decoder module). Additionally or alternatively, iteration 410 can include various other modules for performing other tasks not exemplified in Figure 4 In some examples, refinement network 405 can include a corresponding set of machine learning parameters 485 associated with the machine learning operations. For example, the set of machine learning parameters 485 can be denoted as Θ. This set of machine learning parameters 485 can be set during a machine learning simulation (e.g., a pre-field operation). In some cases, this set of machine learning parameters 485 can include a corresponding parameter for each iteration 410 (e.g., for T iterations, Θ 1 , Θ 2 ,... Θ T where Θ 1 ≠ Θ 2 ≠... ≠ Θ TThis is because the input to each iteration can vary (e.g., different from other traditional encoder and decoder machine learning implementations). In some other cases, the set of machine learning parameters 485 can be common in each iteration 410 to reduce the complexity or size of the implementation of the refinement network 405. That is, the same parameters can be used in each iteration 410 (e.g., θ for T iterations). 1 =θ 2 ,=...=θ T ).

[0134] In some cases, module 425 may output to modules 430 and 435, module 430 may output to module 435, and module 435 may output to the next iteration 410 or to the final (e.g., last, final) output (e.g., output 420) of the refinement network 405. In some examples, the output of module 435 (e.g., output 420) may include channel estimates of the second module 425, the second module 430, and the second module 435 that are output to the next iteration 410 (e.g., a set of MIMO channel estimates for each of the PRG, PRB, and antenna pairs at the respective iteration 410). For example, the set of channel estimates for iteration 410 (e.g., τ) (e.g., H τ ) can be equal to In some cases, the output of module 435 may include latent variables (e.g., a set of latent variables for each channel estimation) that are output to second module 425 and the next iteration 410 of second module 430. For example, for the set of latent variables of iteration 410 (e.g., z... τ ) can be equal to

[0135] In some examples, at least a portion of input 415 may be an input to module 425. For example, as referenced herein... Figure 3 The second set 335 of the multiple channel estimates can be input to module 425. Additionally or alternatively, the observed DMRS symbols (e.g., y...) dmrs ) and known DMRS symbols (e.g., x dmrs ) can be input into module 425 (e.g., (y dmrs ,x dmrs In some cases, module 425 can coordinate the descent of z and H and use cyclic inference as the gradient. For example, according to Equation 3, module 425 can base its measurement on a subset of resources (e.g., DMRS symbol 320) and a second set 335 (H) of multiple channel estimates associated with that subset of resources. τ The difference between x) generates the corresponding set of values ​​for the residual variable 465 (e.g., According to Equation 3, the modulus and the number of mask bits 475 (e.g., binary mask).

[0136] Block 425 can combine respective sets of values of the residual variables 465, the known observations 470 associated with the subset of resources (e.g., ) to generate respective gradients 480 (e.g.,

[0137] Equation 3:

[0138] Accordingly, block 425 can generate respective gradients 480 (e.g., ). For example, each antenna pair can have a respective gradient based on the channel component associated with the respective antenna pair. In some cases, feedback from block 425 (e.g., feedback block) can also be sparse due to the sparsity of the observations.

[0139] In some examples, block 430 can include various steps. For example, block 430 (e.g., encoder block) can receive an input including a first set of values of latent variables, channel estimates, and gradients 480 at 440 (e.g., ); perform fusion of the input at 445 as described herein with reference to Figure 5 ; perform various attention calculations (e.g., intra-PRG calculations at 450-a, inter-PRG calculations at 450-b, and cross-MIMO calculations at 450-c) at 450 as described herein with reference to Figure 6 ; perform a multi-layer perceptron (MLP) process at 455; and output a second set of values of latent variables at 460 (e.g., ). In some cases, the output of each step can be concatenated with the output of the next step. Block 430 can generate the second set of values of latent variables according to Equation 4 via the various steps:

[0140] Equation 4:

[0141] In some cases, block 435 can receive the second set of values of latent variables and modify the channel estimates associated with the plurality of layers based on the second set of values of latent variables, the channel estimates (e.g., channel estimates of a previous iteration), and the respective gradients. For example, the channel estimates of this iteration 410 can be generated according to Equation 5:

[0142] Equation 5:

[0143] as described herein with reference to Figure 5 .

[0144] In some examples, each iteration 410 of refinement network 405 can utilize a same set 485 of one or more machine learning parameters. For example, a first iteration 410 can operate according to this set 485 of machine learning parameters. Each of modules 425, 430, and 435 can operate according to respective machine learning parameters from set 485 of machine learning parameters. Remaining iterations 410 of refinement network 405 can utilize the same set 485 of machine learning parameters to estimate a given attention computation. Set 485 of machine learning parameters can be generated based on training of refinement network 405 according to one or more sets of training parameters. If shared weights (e.g., Θ 1 , Θ 2 ,... Θ T ) are shared during training operations, the same set 485 of machine learning parameters can be generated for each iteration 410. This set 485 of machine learning parameters can be stored on refinement network 405 and can be utilized during operations of refinement network 405. In some examples, different refinement operations can be performed for different attention computations, and all iterations 410 of a given refinement operation can share a respective same set 485 of machine learning parameters, where the set of machine learning parameters for different refinement operations can be different. Attention computations can include, for example, intra-PRB computations, inter-PRB computations, cross-MIMO computations, MLP computations, or any combination thereof. The size of the shared parameter set can be independent of the number of resources and can be based on the number of machine learning model parameters used for a given attention computation.

[0145] Sharing the same set 485 of machine learning parameters across all iterations 410 can enable a reduction in size of refinement network 405 and a reduction in complexity associated with refinement operations compared to refinement operations utilizing different parameters for each iteration 410. The reduction in size and complexity can be obtained while the reliability and accuracy of refinement operations are relatively unchanged. For example, the number of parameters in a set of parameters for iterations 410 of refinement network can be in the range of 100K to 2M parameters, and thus sharing the same set of parameters across iterations 410 can provide substantial benefits in terms of storage space for the set of parameters. Thus, the number of parameters that can be generated when training refinement network 405 can be less than if there are different parameters for each iteration 410, and refinement operations can still produce reliable and accurate outputs 420 associated with refined channel estimates and latent variables.

[0146] Figure 5An example of a network 500 that supports cyclic equivariant inference machine for channel estimation is shown in accordance with one or more aspects of the present disclosure. In some examples, the network 500 can be implemented by aspects of the wireless communications system 100, the network 400, or both. For example, the network 500 can be implemented by a UE 115, a network entity 105, the encoder 430, the decoder 435, or any combination thereof, as described herein with reference to FIGs. 1-4. Figure 1 and 4 .

[0147] In some cases, the network 500 can perform fusion associated with a neural network. For example, the network 500 can be an example of a convolutional neural network (CNN) (e.g., a neural network that uses convolutions instead of more general matrix multiplication for multiple layers). In some cases, the fusion associated with the CNN can fuse (e.g., combine, compress) two or more convolutional layers (e.g., weights associated with each layer) together.

[0148] In some examples, one or more modules (e.g., the module 430, the module 435) of a refinement network (e.g., the refinement network 405) can utilize the network 500, as described herein with reference to Figure 4 For example, an encoder module (e.g., the module 430) can perform a fusion operation (e.g., at 445). The encoder module can generate a second set of values of latent variables associated with the plurality of channel estimates based on respective values of the first set of values of the combined (e.g., fused) channel estimates, respective gradients (e.g., the gradients 480), and the latent variables. For example, the encoder module can generate the output 530 (e.g., ) based on Equation 6:

[0149] Equation 6: where FuCNNis a fusion operation, represents the first set of values of latent variables (e.g., a first portion of the estimates 505), represents the channel estimates (e.g., a second portion of the estimates 505), and represents the respective gradients (e.g., the gradients 510). For example, the encoder module can fuse the gradients 510 (e.g., from the likelihood module 425) into the hidden state variables (e.g., the first set of values of latent variables), thus modeling MIMO multiplexing (e.g., multiplexing MIMO phenomena). The fusion operation can act independently on each PRG (e.g., the PRG 305) of the channel and incorporate the gradients 510 (e.g., gradient information) into the latent state (e.g., perform the fusion operation for each PRG for each antenna pair).

[0150] In some cases, the fusion operation can include various steps. For example, the encoder module can receive estimates 505 (e.g., channel estimates and latent variable values of a previous iteration 410) and gradients 510 (e.g., gradients 480 of the same iteration 410) as inputs, as described herein with reference to Figure 4 The encoder module can combine (e.g., fuse, concatenate) the estimates 505 and the gradients 510 to generate a combination 520 (e.g., ). In some cases, the combination 520 can be input to a u-net 525 (e.g., a tiny u-net). In some examples, the u-net 525 can be an example of a CNN type that utilizes up-sampling operators (e.g., up-sampling operators with a relatively large number of feature channels). The u-net 525 can perform various computations (e.g., combinations, operations) associated with a neural network to generate an output 530 (e.g., ).

[0151] In some examples, the decoder module (e.g., module 435) can perform a fusion operation. The decoder module can modify the channel estimates (e.g., generate a third iteration of channel estimates) based on a second set of values of latent variables (e.g., output from step 460), a second set of channel estimates, and corresponding gradients (e.g., gradients 480). For example, the decoder module can generate the output 530 (e.g., ) based on Equation 7:

[0152] Equation 7: where FuCNNis a fusion operation, denotes the second set of values of latent variables (e.g., a first portion of estimates 505), denotes the channel estimates (e.g., a second portion of estimates 505), and denotes the corresponding gradients (e.g., gradients 510). For example, the decoder module can receive estimates 505 (e.g., channel estimates and latent variable values of the same iteration 410) and gradients 510 (e.g., gradients 480 of the same iteration 410) as inputs, as described herein with reference to Figure 4 The decoder module can combine (e.g., fuse, concatenate) the estimates 505 and the gradients 510 to generate a combination 520 (e.g., ). In some cases, the combination 520 can be input to a u-net 525 (e.g., a tiny u-net). The u-net 525 can perform various computations (e.g., combinations, operations) associated with a neural network to generate an output 530 (e.g., ). The decoder module can utilize information from the various sub-modules (steps 440 through 460) of the likelihood module and the encoder module to update the latent variables, the channel estimates, or both. The decoder module can operate independently on each PRG, utilizing the updated latent variables (e.g., a second set of values) along with the gradient information to improve the channel estimates (e.g., to make the channel estimates closer to the actual channel).

[0153] Figure 6 An example of a network 600 that supports cyclic equivariant inference machine for channel estimation is shown in accordance with one or more aspects of the present disclosure. In some examples, network 600 can be implemented by wireless communication system 100, network 400, or aspects of both. For example, network 600 can be implemented by a UE 115, a network entity 105, encoder 430, or any combination thereof, as described herein with reference to FIGs. 1-4. Figure 1 and 4 described herein.

[0154] In some cases, network 600 can perform attention operations (e.g., attention 605) associated with a neural network. For example, attention 605 can include weighting (e.g., enhancing portions of data while reducing other portions of data) portions of input data (e.g., input 610) differently than other portions of the input data. In some cases, applying attention 605 can aggregate (e.g., modify, align) input data (e.g., observed DMRS symbols) with known data (e.g., known DMRS symbols). In some examples, attention 605 can model interactions (e.g., correlations) within data elements 620. For example, data element 620-a, data element 620-b, data element 620-c, and data element 620-d can interact with data element 620-e, data element 620-f, data element 620-g, and data element 620-h, and vice versa.

[0155] In some cases, an encoder module (e.g., module 430) can perform attention 605 (e.g., self-attention), for example, at various steps (e.g., step 450) of a channel estimation process, as described herein with reference to Figure 4 For example, the encoder module can perform an intra-PRG attention operation (e.g., at 450-a) to model correlations between resources (e.g., resource elements) of each PRB (e.g., PRB 310) of a PRG (e.g., PRG 305) and other PRBs of the PRG (e.g., PRBs belonging to a single PRG bundle). The encoder module can determine a set of values for the latent variables (e.g., output 530) from the fusion operation, as described herein with reference to Figure 5 For example, the encoder module can flatten each subset of values associated with each PRB of the PRG. For example, a set of values (e.g., ) can include four subsets of values (e.g., ). The encoder module can flatten (e.g., combine, compress to a single frequency row) the four subsets into data element 620-a, data element 620-b, data element 620-c, and data element 620-d (e.g., input 610). In some cases, data element 620-e, data element 620-f, data element 620-g, and data element 620-h can be copies (replicas) of data element 620-a, data element 620-b, data element 620-c, and data element 620-d, respectively (e.g., ). In some examples, intra-PRG attention can be used to model long-range correlations (e.g., PRBs separated across the frequency axis).

[0156] In some cases, the encoder module can perform (e.g., at 450-b) an inter-PRG attention operation to model correlations between resources (e.g., resource elements) of each PRG (e.g., PRG 305) of the MIMO communication. The encoder module can determine a flattened subset of values (e.g., ) associated with each PRB of each PRG, and combine (e.g., concatenate, average, mean-pull) the flattened subsets (e.g., embedding subsets) into a single set of values (e.g., ) as data element 620-a, data element 620-b, data element 620-c, and data element 620-d (e.g., input 610), respectively. In some cases, data element 620-e, data element 620-f, data element 620-g, and data element 620-h can be represented by . The encoder module can combine (e.g., add) the respective outputs of attention 605 (e.g., outputs 615, residuals) with the flattened subsets of values (e.g., subsets before averaging ) as output 615 of the inter-PRG attention. In some examples, the inter-PRG attention can facilitate information exchange across different PRG bundles.

[0157] In some cases, the encoder module can perform (e.g., at 450-c) a cross-MIMO attention operation to model correlations between each layer of a plurality of layers associated with the MIMO communication (e.g., interactions between antenna pairs per PRB per PRG bundle). In some cases, the MIMO communication can be a set of resources including DMRS symbols 320 and data symbols 315, as described herein with reference to FIG. 3. The encoder module can determine a source block (e.g., Figure 3 ​). For example, in a two-dimensional grid, data element 620-a, data element 620-b, data element 620-c, and data element 620-d (e.g., inputs 610 for cross-MIMO attention operations) can be represented by and In some cases, data element 620-e, data element 620-f, data element 620-g, and data element 620-h can be represented by and The encoder module can output 615 for cross-MIMO attention. In some examples, the cross-MIMO attention can model interactions (e.g., correlations) between different transmit links per PRB per PRG (e.g., between different MIMO links in an isotropic manner). In some examples, the encoder module can utilize the same set of machine learning parameters for all iterations of a given attention computation. That is, in multiple attention computations, each individual attention computation can utilize a respective set of machine learning parameters, and the set of machine learning parameters can differ across attention computations.

[0158] Figure 7 A block diagram 700 of a device 705 that supports cyclic isotropic reasoning machine for channel estimation in accordance with one or more aspects of the present disclosure is shown. The device 705 can be an example of aspects of a UE 115 or network entity 105 as described herein. The device 705 can include a receiver 710, a transmitter 715, and a communications manager 720. The device 705, or one or more components of the device 705 (e.g., the receiver 710, the transmitter 715, and the communications manager 720), can include at least one processor (or processing circuitry) that can be coupled to at least one memory (or memory circuitry) to individually or collectively support or implement the techniques described herein. Each of these components can be in communication with one another (e.g., via one or more buses).

[0159] The receiver 710 can provide a means for receiving information such as packets, user data, control information, or any combination thereof associated with various information channels (e.g., control channels, data channels, information channels related to cyclic isotropic reasoning machine for channel estimation). Information can be passed on to other components of the device 705. The receiver 710 can utilize a single antenna or a set of multiple antennas.

[0160] The transmitter 715 can provide a means for transmitting signals generated by other components of the device 705. For example, the transmitter 715 can transmit information such as packets, user data, control information, or any combination thereof associated with various information channels (e.g., control channels, data channels, information channels related to cyclic equivariant inference machines for channel estimation, etc.). In some examples, the transmitter 715 can be collocated with the receiver 710 in a transceiver module. The transmitter 715 can utilize a single antenna or a set of multiple antennas.

[0161] The communications manager 720, the receiver 710, the transmitter 715, or various combinations thereof or various components thereof can be examples of means for performing various aspects of cyclic equivariant inference machines for channel estimation as described herein. For example, the communications manager 720, the receiver 710, the transmitter 715, or various combinations or components thereof can be capable of performing one or more of the functions described herein.

[0162] In some examples, the communications manager 720, the receiver 710, the transmitter 715, or various combinations or components thereof can be implemented in hardware (e.g., in communications management circuitry). This circuitry can include a processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described in the present disclosure. In some examples, at least one processor (or processing circuitry) and at least one memory (or memory circuitry) coupled with the at least one processor can be configured to perform one or more of the functions described herein (e.g., executing instructions stored in the at least one memory by the one or more processors, individually or collectively).

[0163] In another implementation, the communications manager 720, the receiver 710, the transmitter 715, or various combinations or components thereof can be implemented in code (e.g., as communications management software or firmware) executed by at least one processor. If implemented in code executed by at least one processor, the functions of the communications manager 720, the receiver 710, the transmitter 715, or various combinations or components thereof can be executed by a general-purpose processor, a DSP, a CPU, an ASIC, an FPGA, a microcontroller, or any combination of these or other programmable logic devices (e.g., configured as or otherwise supporting the means for performing the functions described in the present disclosure).

[0164] In some examples, the communication manager 720 can be configured to use or otherwise employ the receiver 710, the transmitter 715, or both, to perform various operations (e.g., receiving, obtaining, determining, monitoring, outputing, transmitting). For example, the communication manager 720 can receive information from the receiver 710, transmit information to the transmitter 715, or be integrated in combination with the receiver 710, the transmitter 715, or both, to obtain information, output information, or perform various other operations as described herein.

[0165] The communication manager 720 can support wireless communication in accordance with examples as disclosed herein. For example, the communication manager 720 can be configured to, configured to, or operable to support means for receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. The communication manager 720 can be configured to, configured to, or operable to support means for generating, from the reference signal received on the second subset of resources, a first set of channel estimates associated with respective layers of a set of layers of the channel for the set of resources according to MMSE operation. The communication manager 720 can be configured to, configured to, or operable to support means for generating, from a nonlinear two-dimensional interpolation of the channel for the set of resources, a second set of channel estimates and a set of values of a latent variable, the second set of channel estimates and the set of values associated with respective layers of the set of layers of the channel, where the nonlinear two-dimensional interpolation of the channel is based on the first set of channel estimates. The communication manager 720 can be configured to, configured to, or operable to support means for performing a refinement operation on the second set of channel estimates including one or more iterations, where each iteration of the one or more iterations is performed according to a same set of machine learning parameters. In some examples, to perform each iteration of the one or more iterations, the communication manager 720 can be configured to, configured to, or operable to support means for generating a respective gradient associated with the second set of channel estimates based on the second set of channel estimates for the second subset of resources and an observation of a measurement of the second subset of resources, generating a second set of values of the latent variable based on a first set of values of the set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient, and modifying the second set of channel estimates associated with the set of layers based on the second set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient.

[0166] By including or configuring the communications manager 720 in accordance with examples as described herein, the device 705 (e.g., at least one processor of the device 705 controlling or otherwise coupled with the receiver 710, the transmitter 715, the communications manager 720, or a combination thereof) can support a technique for techniques that can support for more accurate channel estimation, more efficient utilization of communication resources, and reduced memory and computational overhead.

[0167] Figure 8 A block diagram 800 of a device 805 that supports cyclic equivariant inference machine for channel estimation in accordance with one or more aspects of the present disclosure is shown. The device 805 can be an example of aspects of a device 705, a UE 115, or a network entity 105 as described herein. The device 805 can include a receiver 810, a transmitter 815, and a communications manager 820. The device 805, or one or more components of the device 805 (e.g., the receiver 810, the transmitter 815, and the communications manager 820), can include at least one processor (or processing circuitry) that can be coupled to at least one memory (or memory circuitry) to support the techniques described herein. Each of these components can be in communication with one another (e.g., via one or more buses).

[0168] The receiver 810 can provide a means for receiving information such as packets, user data, control information, or any combination thereof associated with various information channels (e.g., control channels, data channels, information channels related to cyclic equivariant inference machine for channel estimation). Information can be passed on to other components of the device 805. The receiver 810 can utilize a single antenna or a set of multiple antennas.

[0169] The transmitter 815 can provide a means for transmitting signals generated by other components of the device 805. For example, the transmitter 815 can transmit information associated with various information channels (e.g., control channels, data channels, information channels related to cyclic equivariant inference machine for channel estimation), such as packets, user data, control information, or any combination thereof. In some examples, the transmitter 815 can be collocated with the receiver 810 in a transceiver module. The transmitter 815 can utilize a single antenna or a set of multiple antennas.

[0170] The device 805, or various components thereof, can be an example of means for performing various aspects of cyclic reasoning machine for channel estimation, as described herein. For example, the communication manager 820 can include a scheduling component 825, a MMSE component 830, a coarse network component 835, a refinement network component 840, or any combination thereof. The communication manager 820 can be an example of aspects of the communication manager 720 as described herein. In some examples, the communication manager 820, or various components thereof, can be configured to perform various operations (e.g., receiving, obtaining, monitoring, outputting, transmitting) using or in conjunction with the receiver 810, the transmitter 815, or both. For example, the communication manager 820 can receive information from the receiver 810, transmit information to the transmitter 815, or be integrated in combination with the receiver 810, the transmitter 815, or both, to obtain information, output information, or perform various other operations as described herein.

[0171] The communication manager 820 can support wireless communication in accordance with examples as disclosed herein. The scheduling component 825 can be, be configured as, or be operable to support a means for receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. The MMSE component 830 can be, be configured as, or be operable to support a means for generating, in accordance with MMSE operations, a first set of multiple channel estimates associated with respective layers of a set of multiple layers of the channel for the set of resources from the reference signal received on the second subset of resources. The coarse network component 835 can be, be configured as, or be operable to support a means for generating, in accordance with a nonlinear two-dimensional interpolation of the channel for the set of resources, a second set of multiple channel estimates and a set of multiple values of a latent variable, the second set of multiple channel estimates and the set of multiple values being associated with respective layers of the set of multiple layers of the channel for the set of resources, where the nonlinear two-dimensional interpolation of the channel is based on the first set of multiple channel estimates. The refine network component 840 can be, be configured as, or be operable to support a means for performing a refinement operation on the second set of multiple channel estimates including one or more iterations, where each iteration of the one or more iterations is performed in accordance with a same set of machine learning parameters. In some examples, to perform each iteration of the one or more iterations, the likelihood component 845 can be configured as or otherwise support a means for generating a respective gradient associated with the second set of multiple channel estimates based on the second set of multiple channel estimates for the second subset of resources and an observation of a measurement of the second subset of resources, the encoder component 850 can be configured as or otherwise support a means for generating a second set of values of the latent variable based on a first set of values of the set of multiple values of the latent variable, the second set of multiple channel estimates, the set of machine learning parameters, and the respective gradient, and the decoder component 855 can be configured as or otherwise support a means for modifying the second set of multiple channel estimates associated with the set of multiple layers based on the second set of values of the latent variable, the second set of multiple channel estimates, the set of machine learning parameters, and the respective gradient.

[0172] Additionally, or alternatively, the communication manager 820 can support wireless communication in accordance with examples as disclosed herein. The scheduling component 825 can be configured as or otherwise support a means for receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. The coarse network component 835 can be configured as or otherwise support a means for generating a set of multiple channel estimates associated with respective layers of a set of multiple layers of the channel for the set of resources. The refinement network component 840 can be configured as or otherwise support a means for performing a refinement operation on the set of multiple channel estimates including one or more iterations. In some examples, for each iteration of the one or more iterations, the likelihood component 845 can be configured as or otherwise support a means for generating a respective gradient associated with the set of multiple channel estimates based on the set of multiple channel estimates for the second subset of resources and an observed measurement of the second subset of resources, the encoder component 850 can be configured as or otherwise support a means for generating a second set of values of a latent variable based on a first set of values of the latent variable, the set of multiple channel estimates, and the respective gradient, and the decoder component 855 can be configured as or otherwise support a means for modifying the set of multiple channel estimates associated with the set of multiple layers based on the second set of values of the latent variable, the set of multiple channel estimates, and the respective gradient.

[0173] Figure 9 A block diagram 900 of a communication manager 920 that supports cyclic equivariant inference machine for channel estimation in accordance with one or more aspects of the present disclosure is shown. The communication manager 920 can be an example of aspects of a communication manager 720, a communication manager 820, or both, as described herein. The communication manager 920, or various components thereof, can be an example of means for performing various aspects of a cyclic equivariant inference machine for channel estimation as described herein. For example, the communication manager 920 can include a scheduling component 925, a MMSE component 930, a coarse network component 935, a refinement network component 940, a likelihood component 945, an encoder component 950, a decoder component 955, or any combination thereof. Each of these components, or the components or sub-components thereof, e.g., one or more processors, one or more memories, can communicate, directly or indirectly, with one another (e.g., via one or more buses), the communication can include communications within protocol layers of a protocol stack, communications associated with logical channels of a protocol stack (e.g., between protocol layers of the protocol stack, within a device, component, or virtualized component associated with the network entity 105, between devices, components, or virtualized components associated with the network entity 105), or any combination thereof.

[0174] The communication manager 920 can support wireless communication in accordance with examples as disclosed herein. The scheduling component 925 can enable, be configured as, or be operable to support means for receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. The MMSE component 930 can enable, be configured as, or be operable to support means for generating, from the reference signal received on the second subset of resources, a first set of channel estimates associated with respective layers of a set of layers of the channel for the set of resources in accordance with MMSE operations. The coarse network component 935 can enable, be configured as, or be operable to support means for generating, from a nonlinear two-dimensional interpolation of the channel for the set of resources, a second set of channel estimates and a set of values of a latent variable, the second set of channel estimates and the set of values associated with respective layers of the set of layers of the channel, where the nonlinear two-dimensional interpolation of the channel is based on the first set of channel estimates. The refinement network component 940 can enable, be configured as, or be operable to support means for performing a refinement operation on the second set of channel estimates including one or more iterations, where each iteration of the one or more iterations is performed in accordance with a same set of machine learning parameters. In some examples, to perform each iteration of the one or more iterations, the likelihood component 945 can enable, be configured as, or be operable to support means for generating a respective gradient associated with the second set of channel estimates based on the second set of channel estimates for the second subset of resources and an observation of a measurement of the second subset of resources, the encoder component 950 can enable, be configured as, or be operable to support means for generating a second set of values of the latent variable based on a first set of values of the set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient, and the decoder component 955 can enable, be configured as, or be operable to support means for modifying the second set of channel estimates associated with the set of layers based on the second set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient.

[0175] In some examples, the refinement operation is a first refinement operation and the set of machine learning parameters is a first set of machine learning parameters, and the refinement network component 940 can be, configured to, or operable to support a means for performing a second refinement operation on a second set of multiple channel estimates, the second refinement operation including one or more second iterations performed according to a same second set of machine learning parameters, where the first refinement operation and the second refinement operation are associated with respective attention computations of the set of multiple attention computations.

[0176] In some examples, the set of multiple attention computations includes intra-PRB group computations, inter-PRB group computations, cross-MIMO computations, MLP computations, or any combination thereof.

[0177] In some examples, the MMSE component 930 can be, configured to, or operable to support a means for performing the MMSE operation based on a resource configuration pattern of the second subset of resources allocated for the reference signal, the reference signal including a DMRS.

[0178] In some examples, to support generating the respective gradient, the likelihood component 945 can be, configured to, or operable to support a means for generating a respective set of values of residual variables based on a difference between the measured observation for the second subset of resources and the second set of multiple channel estimates of the second subset of resources. In some examples, to support generating the respective gradient, the likelihood component 945 can be, configured to, or operable to support a means for combining the respective set of values of residual variables, the second subset of resources, and a number of mask bits.

[0179] In some examples, to support generating the second set of values of latent variables, the encoder component 950 can be, configured to, or operable to support a means for combining, based on generating the respective gradient, the second set of multiple channel estimates for the second subset of resources, the respective gradient, and a respective value of the first set of values of latent variables.

[0180] In some examples, to support generating the second set of values of latent variables, the encoder component 950 can be, configured to, or operable to support a means for modeling a correlation between a resource of each resource block in a resource block group and other resource blocks in the resource block group.

[0181] In some examples, to support generating the second set of values of the latent variable, the encoder component 950 can be, configured as, or operable with, means for modeling a correlation between resources of each group of a set of multiple groups of resources of the resource set and other groups of the set of multiple groups of resources, where each group of the set of multiple groups of resources includes a set of multiple resource blocks.

[0182] In some examples, to support generating the second set of values of the latent variable, the encoder component 950 can be, configured as, or operable with, means for modeling a correlation between each layer of the set of multiple layers for the resource set.

[0183] In some examples, to support modifying the second set of multiple channel estimates, the encoder component 950 can be, configured as, or operable with, means for combining the second set of values of the latent variable, the second set of multiple channel estimates, and the respective gradient based on the set of machine learning parameters.

[0184] In some examples, the non-linear two-dimensional interpolation of the channel is based on a machine learning model.

[0185] In some examples, the first set of multiple channel estimates and the second set of multiple channel estimates are associated with a set of multiple single-input and single-output antenna pairs.

[0186] In some examples, each iteration of the one or more iterations is performed by a refinement network including a likelihood module, an encoder module, and a decoder module, the refinement network including a machine learning model. In some examples, each refinement network is performed according to the same set of machine learning parameters.

[0187] The communication manager 920 can support wireless communication in accordance with examples as disclosed herein. The scheduling component 925 can be configured as or otherwise support a means for receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. The coarse network component 935 can be configured as or otherwise support a means for generating a set of multiple channel estimates associated with respective layers of a set of multiple layers of the channel for the set of resources. The refinement network component 940 can be configured as or otherwise support a means for performing a refinement operation on the set of multiple channel estimates including one or more iterations. In some examples, for each iteration of the one or more iterations, the likelihood component 945 can be configured as or otherwise support a means for generating a respective gradient associated with the set of multiple channel estimates based on the set of multiple channel estimates for the second subset of resources and an observed measurement of the second subset of resources, an encoder component 950 can be configured as or otherwise support a means for generating a second set of values of a latent variable based on a first set of values of the latent variable, the set of multiple channel estimates, and the respective gradient, and a decoder component 955 can be configured as or otherwise support a means for modifying the set of multiple channel estimates associated with the set of multiple layers based on the second set of values of the latent variable, the set of multiple channel estimates, and the respective gradient.

[0188] In some examples, to support generating the respective gradient, the likelihood component 945 can be configured as or otherwise support a means for generating a respective set of values of a residual variable based on a difference between the observed measurement of the second subset of resources and the set of multiple channel estimates for the second subset of resources. In some examples, to support generating the respective gradient, the likelihood component 945 can be configured as or otherwise support a means for combining the respective set of values of the residual variable, the observed measurement of the second subset of resources, and a number of mask bits.

[0189] In some examples, to support generating the second set of values of the latent variable, the encoder component 950 can be configured as or otherwise support a means for combining, based on generating the respective gradient, the set of multiple channel estimates for the second subset of resources, the respective gradient, and a respective value of the first set of values of the latent variable.

[0190] In some examples, to support generating the second set of values of the latent variable, the encoder component 950 can be configured as or otherwise support a means for modeling a correlation between resources of each resource block of a group of resource blocks and other resource blocks of the group of resource blocks.

[0191] In some examples, to support generating the second set of values of the latent variable, the encoder component 950 can be configured as or otherwise support a means for modeling a correlation between resources of each group of a set of multiple groups of resources of the resource set and other groups of the set of multiple groups of resources, where each group of the set of multiple groups of resources includes a set of multiple resource blocks.

[0192] In some examples, to support generating the second set of values of the latent variable, the encoder component 950 can be configured as or otherwise support a means for modeling a correlation between each layer of the set of multiple layers for the resource set.

[0193] In some examples, to support modifying the set of multiple channel estimates, the decoder component 955 can be configured as or otherwise support a means for combining the set of values of the latent variable, the second set of multiple channel estimates, and the respective gradient.

[0194] In some examples, the initial values of the set of multiple channel estimates are associated with SISO antenna pairs.

[0195] In some examples, the second subset of resources is configured according to a resource configuration pattern of a set of resource configuration patterns.

[0196] In some examples, the set of resource configuration patterns is a set of DMRS patterns.

[0197] In some examples, each iteration is performed by a refinement network that includes a likelihood module, an encoder module, and a decoder module, and each refinement network further includes respective parameters associated with machine learning operations.

[0198] In some examples, the resource set includes one or more groups of resources, and each respective layer of the set of multiple layers is associated with a respective antenna pair of a set of multiple SISO antenna pairs.

[0199] Figure 10A diagram of a system 1000 including a device 1005 that supports cyclic equivariant reasoning machine for channel estimation in accordance with one or more aspects of the present disclosure is shown. The device 1005 can be an example of or include the components of device 705, device 805, or a UE 115 as described herein. The device 1005 can communicate with one or more network entities 105, one or more UEs 115, or any combination thereof (e.g., wirelessly). The device 1005 can include components for bi-directional voice and data communications including components for transmitting and receiving communications, such as a communications manager 1020, an input / output (I / O) controller 1010, a transceiver 1015, an antenna 1025, at least one memory 1030 (or memory circuitry), code 1035, and at least one processor 1040 (or processing circuitry). These components can be in electronic communication or otherwise coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more buses (e.g., bus 1045).

[0200] The I / O controller 1010 can manage input and output signals for the device 1005. The I / O controller 1010 can also manage peripherals not integrated into the device 1005. In some cases, the I / O controller 1010 can represent a physical connection or port to the external peripherals. In some cases, the I / O controller 1010 can utilize an operating system such as iOS®, ANDROID®, MS-DOS®, MS-WINDOWS®, OS / 2®, LINUX®, or another known operating system. Additionally or alternatively, the I / O controller 1010 can represent or interact with a modem, a keyboard, a mouse, a touchscreen, or similar devices or In some cases, the I / O controller 1010 can be implemented as part of one or more processors (such as the at least one processor 1040) (or processing circuitry). In some cases, a user can interact with the device 1005 via the I / O controller 1010 or via hardware components controlled by the I / O controller 1010.

[0201] In some cases, the device 1005 can include a single antenna 1025. However, in some other cases the device 1005 can have more than one antenna 1025, which can be capable of concurrently sending or receiving multiple wireless transmissions. The transceiver 1015 can communicate bi-directionally, via the one or more antennas 1025, wired, or wireless links as described herein. For example, the transceiver 1015 can represent a wireless transceiver and can communicate bi-directionally with another wireless transceiver. The transceiver 1015 can also include a modem to modulate the packets and to demodulate packets received from one or more antennas 1025. The transceiver 1015, either alone or in combination with one or more antennas 1025, can be an example of a transmitter 715, a transmitter 815, a receiver 710, a receiver 810, or any combination thereof, or a component thereof, as described herein.

[0202] The at least one memory 1030 (or memory circuitry) can include random access memory (RAM) and read-only memory (ROM). The at least one memory 1030 can store computer-readable, computer-executable code 1035 including instructions that, when executed by the at least one processor 1040, cause the device 1005 to perform various functions described herein. The code 1035 can be stored in a non-transitory computer-readable medium such as system memory or another type of memory. In some cases, the code 1035 can not be directly executable by the at least one processor 1040 but can cause a computer (e.g., when compiled and executed) to perform functions described herein. In some cases, the at least one memory 1030 can include, among other things, a basic I / O system (BIOS), which can control basic hardware or software operation such as the interaction with peripheral components or devices.

[0203] The at least one processor 1040 (or processing circuitry) can include an intelligent hardware device, e.g., a general- purpose processor, a DSP, a CPU, a microcontroller, an ASIC, an FPGA, a programmable logic device, a discrete gate or transistor logic component, a discrete hardware component, or any combination thereof. In some cases, the at least one processor 1040 can be configured to operate a memory array. In some other cases, a memory controller can be integrated into the at least one processor 1040. The at least one processor 1040 can be configured to execute computer-readable instructions stored in a memory (e.g., the at least one memory 1030) to cause the device 1005 to perform various functions (e.g., functions or tasks supporting a cyclic equivariant inference machine for channel estimation). For example, the device 1005 or a component of the device 1005 can include the at least one processor 1040 (or processing circuitry) and the at least one memory 1030 (or memory circuitry) coupled with or to the at least one processor 1040 and configured to perform various functions described herein. In some examples, the at least one processor 1040 can include a plurality of processors, and the at least one memory 1030 can include a plurality of memories. One or more processors of the plurality of processors can be coupled with one or more memories of the plurality of memories, which can individually or collectively be configured to perform the various functions described herein. In some examples, the at least one processor 1040 can be a component of a processing system, which can refer to a machine (such as a series of machines), a circuit (including, for example, one or both of a processor circuit (which can include the at least one processor 1040) and a memory circuit (which can include the at least one memory 1030)), or a system of components that receives or obtains inputs and processes the inputs to produce, generate, or obtain a set of outputs. The processing system can be configured to perform one or more of the functions described herein. Thus, the at least one processor 1040, or a processing system that includes the at least one processor 1040, can be configured to, can be configurable to, or can be operable to cause the device 1005 to perform one or more of the functions described herein. Moreover, as described herein, “configured to,” “configurable to,” and “operable to” can be used interchangeably and can be associated with the ability of the at least one processor 1040, or a processing system that includes the at least one processor 1040, to perform one or more of the functions described herein when executing code stored in the at least one memory 1030 or otherwise executing the functions.

[0204] The communications manager 1020 can support wireless communication in accordance with examples as disclosed herein. For example, the communications manager 1020 can be, be configured as, or can operate a means for receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. The communications manager 1020 can be, be configured as, or can operate a means for generating, from the reference signal received on the second subset of resources, a first set of channel estimates associated with respective layers of a set of layers of the channel for the set of resources according to MMSE operation. The communications manager 1020 can be, be configured as, or can operate a means for generating, from a nonlinear two-dimensional interpolation of the channel for the set of resources, a second set of channel estimates and a set of values of a latent variable, the second set of channel estimates and the set of values associated with respective layers of the set of layers of the channel for the set of resources, where the nonlinear two-dimensional interpolation of the channel is based on the first set of channel estimates. The communications manager 1020 can be, be configured as, or can operate a means for performing a refinement operation on the second set of channel estimates including one or more iterations, where each iteration of the one or more iterations is performed according to a same set of machine learning parameters. In some examples, to perform each iteration of the one or more iterations, the communications manager 1020 can be configured as, or can otherwise support, a means for generating a respective gradient associated with the second set of channel estimates based on the second set of channel estimates for the second subset of resources and an observation of a measurement of the second subset of resources, a means for generating a second set of values of the latent variable based on a first set of values of the set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient, and a means for modifying the second set of channel estimates associated with the set of layers based on the second set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient.

[0205] Additionally, or alternatively, the communication manager 1020 can support wireless communication in accordance with examples as disclosed herein. For example, the communication manager 1020 can be configured as or otherwise support a means for receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. The communication manager 1020 can be configured as or otherwise support a means for generating a set of multiple channel estimates associated with respective layers of a set of multiple layers of the channel for the set of resources. The communication manager 1020 can be configured as or otherwise support a means for performing a refinement operation on the set of multiple channel estimates including one or more iterations. In some examples, for each iteration of the one or more iterations, the communication manager 1020 can be configured as or otherwise support a means for generating a respective gradient associated with the set of multiple channel estimates based on the set of multiple channel estimates for the second subset of resources and an observed measurement of the second subset of resources, generating a second set of values of a latent variable based on a first set of values of the latent variable, the set of multiple channel estimates, and the respective gradient, and modifying the set of multiple channel estimates associated with the set of multiple layers based on the second set of values of the latent variable, the set of multiple channel estimates, and the respective gradient.

[0206] By including or configuring the communication manager 1020 in accordance with examples as described herein, the device 1005 can support techniques for improving communication reliability, reducing latency, improving user experience related to reduced processing, reducing power consumption, more efficient utilization of communication resources, more accurate channel estimates, and reduced memory and computational overhead.

[0207] In some examples, the communication manager 1020 can be configured to use or otherwise coordinate with the transceiver 1015, the one or more antennas 1025, or any combination thereof, in performing various operations (e.g., receiving, monitoring, transmitting) described herein. Although the communication manager 1020 is illustrated as a separate component, in some examples, one or more functions described with reference to the communication manager 1020 can be supported by, or performed by, the at least one processor 1040, the at least one memory 1030, the code 1035, or any combination thereof. For example, the code 1035 can include instructions executable by the at least one processor 1040 to cause the device 1005 to perform various aspects of a variational inference machine for channel estimation as described herein, or the at least one processor 1040 and the at least one memory 1030 can be otherwise configured to individually or collectively perform or support performance of such operations.

[0208] Figure 11A diagram illustrating a system 1100 including a device 1105 that supports cyclic equivariant reasoning machine for channel estimation in accordance with one or more aspects of the present disclosure is shown. The device 1105 can be an example of or include the components of device 705, device 805, or a network entity 105 as described herein. The device 1105 can communicate with one or more network entities 105, one or more UEs 115, or any combination thereof, the communication can include communications through one or more wired interfaces, through one or more wireless interfaces, or any combination thereof. The device 1105 can include components for supporting output and obtaining communications, such as a communication manager 1120, a transceiver 1110, an antenna 1115, at least one memory 1125 (or memory circuitry), code 1130, and at least one processor 1135 (or processing circuitry). These components can be in electronic communication or otherwise coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more buses (e.g., bus 1140).

[0209] The transceiver 1110 can support bi-directional communication over a wired link, a wireless link, or both, as described herein. In some examples, the transceiver 1110 can include a wired transceiver and can communicate bi-directionally with another wired transceiver. Additionally or alternatively, in some examples, the transceiver 1110 can include a wireless transceiver and can communicate bi-directionally with another wireless transceiver. In some examples, the device 1105 can include one or more antennas 1115, which can be capable of transmitting or receiving wireless transmissions (e.g., concurrently). The transceiver 1110 can also include a modem to modulate signals; provide the modulated signals to the one or more antennas 1115 for transmission (e.g., by a transmitter), receive the modulated signals (e.g., from the one or more antennas 1115, from a receiver), and demodulate the signals. In some implementations, the transceiver 1110 can include one or more interfaces, such as one or more interfaces coupled with the one or more antennas 1115 configured to support various receive or obtain operations, or one or more interfaces coupled with the one or more antennas 1115 configured to support various transmit or output operations, or a combination thereof. In some implementations, the transceiver 1110 can include or be coupled with one or more processors or one or more memory components capable of operating to perform or support operations based on received or obtained information or signals, or to generate information or other signals for transmission or other output, or any combination thereof. In some implementations, the transceiver 1110, or the transceiver 1110 and the one or more antennas 1115, or the transceiver 1110 and the one or more antennas 1115 and the one or more processors or the one or more memory components (e.g., the at least one processor 1135, the at least one memory 1125, or both) can be included in a chip or chip assembly mounted in the device 1105. In some examples, the transceiver 1110 can operate to support communication via one or more communication links (e.g., the communication links 125, the backhaul communication links 120, the in-transit communication links 162, the front-haul communication links 168).

[0210] The at least one memory 1125 can include RAM, ROM, or any combination thereof. The at least one memory 1125 can store computer-readable, computer-executable code 1130 including instructions that, when executed by the one or more processors of the at least one processor 1135, cause the device 1105 to perform various functions described herein. The code 1130 can be stored in a non-transitory computer-readable medium such as system memory or another type of memory. In some cases, the code 1130 can not be directly executable by the processor(s) of the at least one processor 1135 but can cause a computer (e.g., when compiled and executed) to perform functions described herein. In some cases, the at least one memory 1125 can include, among other things, a BIOS which can control basic hardware or software operation such as the interaction with peripheral components or devices. In some examples, the at least one processor 1135 can include a plurality of processors, and the at least one memory 1125 can include a plurality of memories. One or more processors of the plurality of processors can be coupled to one or more memories of the plurality of memories, which can be configured individually or collectively to perform various functions (e.g., as part of a processing system) described herein.

[0211] The at least one processor 1135 can include an intelligent hardware device, e.g., a general- purpose processor, a DSP, an ASIC, a CPU, a FPGA, a microcontroller, a programmable logic device, a discrete gate or transistor logic component, a discrete hardware component, or any combination thereof. In some cases, the at least one processor 1135 can be configured to operate a memory array using a memory controller. In some other cases, a memory controller can be integrated into one or more processors of the at least one processor 1135. The at least one processor 1135 can be configured to execute computer-readable instructions stored in a memory (e.g., the at least one memory 1125 or other memory) to cause the device 1105 to perform various functions (e.g., functions or tasks supporting a cyclic equivariant inference machine for channel estimation). For example, the device 1105 or a component of the device 1105 can include the at least one processor 1135 (or processing circuitry) and the at least one memory 1125 (or memory circuitry) coupled with or to one or more of the at least one processor 1135, which can be configured to perform various functions described herein. The at least one processor 1135 can be an example of a cloud computing platform (e.g., one or more physical nodes and supporting software such as an operating system, virtual machines, or container instances) that can host functions for performing functions of the device 1105 (e.g., by executing code 1130). The at least one processor 1135 can be any one or more suitable processors capable of executing scripts or instructions of one or more software programs stored in the device 1105, such as within one or more memories of the at least one memory 1125. In some examples, the at least one processor 1135 can include multiple processors, and the at least one memory 1125 can include multiple memories. One or more of the multiple processors can be coupled with one or more of the multiple memories, which can be individually or collectively configured to perform various functions described herein. In some examples, the at least one processor 1135 can be a component of a processing system, which can refer to a machine (such as a series of machines), a system of circuits (including, for example, one or both of processor circuitry (which can include the at least one processor 1135) and memory circuitry (which can include the at least one memory 1125)), or a system of components that receives or obtains inputs and processes the inputs to produce, generate, or obtain a set of outputs. The processing system can be configured to perform one or more of the functions described herein. Thus, the at least one processor 1135 or a processing system including the at least one processor 1135 can be configured to, can be configured to, or can be operable to cause the device 1105 to perform one or more of the functions described herein.Moreover, as described herein, "configured to," "configurable to," and "operable to" are used interchangeably and can be associated with the ability of the at least one processor 1135, in execution of code stored in the at least one memory 1125, or otherwise, to perform one or more of the functions described herein.

[0212] In some examples, the bus 1140 can support communication of protocol layers (e.g., within protocol layers) of a protocol stack. In some examples, the bus 1140 can support communication associated with logical channels of a protocol stack (e.g., between protocol layers of a protocol stack), which can include communication performed within a component of the device 1105, or between different components of the device 1105 that can be co-located or located at different locations (e.g., where the device 1105 can refer to a system in which one or more of the communication manager 1120, transceiver 1110, at least one memory 1125, code 1130, and at least one processor 1135 can be located in one component or partitioned between different components).

[0213] In some examples, the communication manager 1120 can manage aspects of communication with the core network 130 (e.g., via one or more wired or wireless backhaul links). For example, the communication manager 1120 can manage the transfer of data communications for client devices, such as one or more UEs 115. In some examples, the communication manager 1120 can manage communications with other network entities 105, and can include a controller or scheduler for coordinating communications with UEs 115 in cooperation with other network entities 105. In some examples, the communication manager 1120 can support an X2 interface within an LTE / LTE-A wireless communication network technology to provide communication between network entities 105.

[0214] The communications manager 1120 can support wireless communication as described herein. For example, the communications manager 1120 can be configured as or operate as a means for receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. The communications manager 1120 can be configured as or operate as a means for generating, according to MMSE operations, a first set of multiple channel estimates associated with respective layers of a set of multiple layers of the channel for the set of resources from the reference signal received on the second subset of resources. The communications manager 1120 can be configured as or operate as a means for generating, according to a nonlinear two-dimensional interpolation of the channel for the set of resources, a second set of multiple channel estimates and a set of multiple values of a latent variable, the second set of multiple channel estimates and the set of multiple values associated with respective layers of the set of multiple layers of the channel, where the nonlinear two-dimensional interpolation of the channel is based on the first set of multiple channel estimates. The communications manager 1120 can be configured as or operate as a means for performing a refinement operation on the second set of multiple channel estimates including one or more iterations, where each iteration of the one or more iterations is performed according to a same set of machine learning parameters. In some examples, to perform each iteration of the one or more iterations, the communications manager 1120 can be configured as or otherwise support a means for generating a respective gradient associated with the second set of multiple channel estimates based on the second set of multiple channel estimates for the second subset of resources and observations of measurements of the second subset of resources, generating a second set of values of the latent variable based on a first set of values of the set of multiple values of the latent variable, the second set of multiple channel estimates, the set of machine learning parameters, and the respective gradient, and modifying the second set of multiple channel estimates associated with the set of multiple layers based on the second set of values of the latent variable, the second set of multiple channel estimates, the set of machine learning parameters, and the respective gradient.

[0215] By including or configuring the communications manager 1120 in accordance with examples described herein, the device 1105 can support a technique for improving communication reliability, reducing latency, improving user experience related to processing, reducing power consumption, more efficient utilization of communication resources, more accurate channel estimation, and reducing memory and computational overhead.

[0216] In some examples, the communication manager 1120 can be configured to use or otherwise employ the transceiver 1110, the one or more antennas 1115 (e.g., where applicable), or any combination thereof, to perform various ones of the operations (e.g., receiving, obtaining, monitoring, outputting, sending). Although the communication manager 1120 is illustrated as a separate component, in some examples, one or more functions described with reference to the communication manager 1120 can be supported by, or performed by, the transceiver 1110, one or more of the at least one processor 1135, one or more of the at least one memory 1125, the code 1130, or any combination thereof (e.g., by a processing system including at least a portion of the at least one processor 1135, the at least one memory 1125, the code 1130, or any combination thereof). For example, the code 1130 can include instructions executable by one or more of the at least one processor 1135 to cause the device 1105 to perform various aspects of a cyclic equivariant machine for channel estimation as described herein, or the at least one processor 1135 and the at least one memory 1125 can be otherwise configured to individually or collectively perform or support performance of such operations.

[0217] Figure 12 A flow diagram illustrating a method 1200 that supports a cyclic equivariant machine for channel estimation in accordance with aspects of the present disclosure is shown. The operations of method 1200 can be implemented by a UE or network entity or its components as described herein. For example, the operations of method 1200 can be performed by a UE 115 or network entity as described with reference to FIGs. 1 through 3 and 5 through 7. In some examples, a UE or network entity can execute a set of instructions to control the functional elements of the UE or network entity to perform the described functions. Additionally or alternatively, the UE or network entity can perform aspects of the described functions using special-purpose hardware. Figures 1 to 11

[0218] At 1205, the method can include receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. The operations of block 1205 can be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1205 can be performed by a scheduling component 925 as described with reference to FIGs. 1 through 3 and 5 through 7. Figure 9

[0219] At 1210, the method can include generating, according to a MMSE operation, a first set of channel estimates associated with respective layers of a set of layers of the channel for the set of resources from the reference signal received on the second subset of resources. The operations of block 1210 can be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1210 can be performed by a channel estimation component 930 as described with reference to FIGs. 1 through 3 and 5 through 7. Figure 9 ​​The MMSE component 930 described above can perform the operations.

[0220] At 1215, the method can include generating a second set of channel estimates and a set of values of latent variables according to a non-linear two-dimensional interpolation of the channel for the set of resources, the second set of channel estimates and the set of values associated with respective layers of the set of layers of the channel for the set of resources, where the non-linear two-dimensional interpolation of the channel is based on the first set of channel estimates. The operations of block 1215 can be performed according to the methods described herein. In some examples, aspects of the operations of 1215 can be performed by a non-linear two-dimensional interpolation component as described with reference to Figure 9 The coarse network component 935 described above can perform the operations.

[0221] At 1220, the method can include performing a refinement operation on the second set of channel estimates including one or more iterations, where each iteration of the one or more iterations is performed according to the same set of machine learning parameters. In some examples, each iteration of the one or more iterations can include generating a respective gradient associated with the second set of channel estimates based on the second set of channel estimates for the second subset of resources and observations of measurements of the second subset of resources, generating a second set of values of the latent variable based on a first set of values of the set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient, and modifying the second set of channel estimates associated with the set of layers based on the second set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient. The operations of block 1220 can be performed according to the methods described herein. In some examples, aspects of the operations of 1220 can be performed by a refinement network component as described with reference to Figure 9 The refinement network component 940 described above can perform the operations.

[0222] Figure 13 A flow diagram illustrating a method 1300 that supports cyclic equivariant reasoning machine for channel estimation in accordance with aspects of the present disclosure is shown. The operations of method 1300 can be implemented by a UE or network entity or its components as described herein. For example, the operations of method 1300 can be performed by a UE 115 or network entity as described with reference to Figures 1 to 11 In some examples, a UE or network entity can execute a set of instructions to control the functional elements of the UE or network entity to perform the described functions. Additionally or alternatively, the UE or network entity can perform aspects of the described functions using special-purpose hardware.

[0223] At 1305, the method can include receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. The operations of block 1305 can be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1305 can be performed by a scheduling component 925 as described with reference to Figure 9 FIG. 9.

[0224] At 1310, the method can include generating, from the reference signal received on the second subset of resources, a first set of channel estimates associated with respective layers of a set of layers of the channel for the set of resources according to MMSE operations. The operations of block 1310 can be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1310 can be performed by a MMSE component 930 as described with reference to Figure 9 FIG. 9.

[0225] At 1315, the method can include generating a second set of channel estimates and a set of values of a latent variable associated with respective layers of the set of layers of the channel for the set of resources according to a nonlinear two-dimensional interpolation of the channel for the set of resources, where the nonlinear two-dimensional interpolation of the channel is based on the first set of channel estimates. The operations of block 1315 can be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1315 can be performed by a coarse network component 935 as described with reference to Figure 9 FIG. 9.

[0226] At 1320, the method can include performing a refinement operation on the second set of channel estimates including one or more iterations, where each iteration of the one or more iterations is performed according to a same set of machine learning parameters. In some examples, each iteration of the one or more iterations can include generating a respective gradient associated with the second set of channel estimates based on the second set of channel estimates for the second subset of resources and an observation of a measurement of the second subset of resources, generating a second set of values of the latent variable based on a first set of values of the set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient, and modifying the second set of channel estimates associated with the set of layers based on the second set of values of the latent variable, the second set of channel estimates, the set of machine learning parameters, and the respective gradient. The operations of block 1320 can be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1320 can be performed by a refinement network component 940 as described with reference to Figure 9 FIG. 9.

[0227] At 1325, the method can include performing a second refinement operation on the second set of multiple channel estimates, the second refinement operation including one or more second iterations performed according to a same second set of machine learning parameters, where the first refinement operation and the second refinement operation are associated with respective attention computations of the set of multiple attention computations. The operations of block 1325 can be performed according to the examples as disclosed herein. In some examples, aspects of the operations of 1325 can be performed by a refinement network component 940 as described with reference to Figure 9 The operations of block 1325 can be performed according to the examples as disclosed herein. In some examples, aspects of the operations of 1325 can be performed by a refinement network component 940 as described with reference to

[0228] Figure 14 A flow diagram illustrating a method 1400 that supports cyclic equivariant reasoning machine for channel estimation in accordance with one or more aspects of the present disclosure is shown. The operations of method 1400 can be implemented by a UE or its components as described herein. For example, the operations of method 1400 can be performed by a UE 115 as described with reference to Figures 1 to 13 FIGS. 13 through 15, and / or any other suitable component of a wireless

[0229] At 1405, the method can include receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. Receiving the assignment can include identifying time-frequency resources on which the assignment is transmitted, demodulating a transmission over those time-frequency resources, and decoding the demodulated transmission to obtain bits indicative of the assignment. The assignment can be received in a downlink control channel via DCI. The operations of 1405 can be performed according to the examples as disclosed herein. In some examples, aspects of the operations of 1405 can be performed by a scheduling component 925 as described with reference to Figure 9 FIGS. 13 through 15, and / or any other suitable component of a wireless

[0230] At 1410, the method can include generating a set of multiple channel estimates associated with respective layers of a set of multiple layers of the channel for the set of resources. Generating the set of multiple channel estimates can include performing various interpolation techniques as described herein with reference to Figure 3 The operations of 1410 can be performed according to the examples as disclosed herein. In some examples, aspects of the operations of 1410 can be performed by a coarse network component 935 as described with reference to Figure 9 FIGS. 13 through 15, and / or any other suitable component of a wireless

[0231] At 1415, the method can include performing a refinement operation on the set of multiple channel estimates including one or more iterations. Performing the refinement operation can include via a refinement network component 940 as described herein with reference to Figures 4 to 6The refinement network computes (e.g., updates, generates) various channel estimates (e.g., MIMO channel estimates) and refines the estimates over various iterations of machine learning operations. In some examples, each iteration of the one or more iterations can include generating a respective gradient associated with the set of multiple channel estimates based on the set of multiple channel estimates for the second subset of resources and the observed measurements of the second subset of resources, generating a second set of values of the latent variable based on the first set of values of the latent variable, the set of multiple channel estimates, and the respective gradient, and modifying the set of multiple channel estimates associated with the set of multiple layers based on the second set of values of the latent variable, the set of multiple channel estimates, and the respective gradient. The operations of 1415 can be performed according to the examples as disclosed herein. In some examples, aspects of the operations of 1415 can be performed by a refinement network component 940 as described with reference to Figure 9 The operations of 1415 can be performed according to the examples as disclosed herein. In some examples, aspects of the operations of 1415 can be performed by a refinement network component 940 as described with reference to

[0232] Figure 15 A method 1500 that supports cyclic equivariant reasoning machine for channel estimation is illustrated. The operations of method 1500 can be implemented by a UE or its components as described herein. For example, the operations of method 1500 can be performed by a UE 115 as described with reference to Figures 1 to 14 FIG. 15 illustrates examples of method 1500 that support cyclic equivariant reasoning machine for channel estimation in accordance with aspects of the present disclosure. The operations of method 1500 can be implemented by a UE or its components as described herein. For example, the operations of method 1500 can be performed by a UE 115 as described with reference to

[0233] At 1505, the method can include receiving an assignment of a set of resources associated with a channel, the set of resources including a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal. Receiving the assignment can include identifying time-frequency resources on which the assignment is transmitted, demodulating a transmission over those time-frequency resources, decoding the demodulated transmission to obtain bits indicative of the assignment. The assignment can be received in a downlink control channel via DCI. The operations of 1505 can be performed according to the examples as disclosed herein. In some examples, aspects of the operations of 1505 can be performed by a scheduling component 925 as described with reference to Figure 9 The operations of 1505 can be performed according to the examples as disclosed herein. In some examples, aspects of the operations of 1505 can be performed by a scheduling component 925 as described with reference to

[0234] At 1510, the method can include generating a set of multiple channel estimates associated with a respective layer of a set of multiple layers of a channel for the set of resources. Generating the set of multiple channel estimates can include performing various interpolation techniques as described herein with reference to Figure 3 The operations of 1510 can be performed according to the examples as disclosed herein. In some examples, aspects of the operations of 1510 can be performed by a refinement network component 940 as described with reference toFigures 3 to 5 The coarse network component 935 is used to perform this.

[0235] At 1515, the method may include performing a refinement operation on the set of multiple channel estimates comprising one or more iterations. Performing the refinement operation may include via means as referenced herein. Figure 9 The refinement network computation (e.g., updating, generating) of various channel estimates (e.g., MIMO channel estimates) and refinement of the estimates over various iterations of the machine learning operation. In some examples, each of the one or more iterations may include: generating a corresponding gradient associated with the set of multiple channel estimates based on observations of measurements of the second resource subset and a set of multiple channel estimates for the second resource subset; generating a second set of latent variable values ​​(e.g., z) based at least in part on a first set of latent variable values, the multiple channel estimates, the corresponding gradients, and modeling correlations between resources in each of the multiple resource groups of the resource set and other groups in the multiple resource groups. τ The set of multiple resource groups includes multiple resource blocks; and the set of multiple channel estimates associated with the set of multiple layers is modified based on the second set of the latent variable values, the set of multiple channel estimates, and the corresponding gradient. The operation of 1515 can be performed according to examples disclosed herein. In some examples, aspects of the operation of 1515 can be performed by a likelihood module (e.g., module 425), an encoder module (e.g., module 430), or a decoder module (e.g., module 435). In some examples, aspects of the operation of 1515 can be performed by, as referenced... ​ The refined network component 940 is used to perform this.

[0236] The following provides an overview of the various aspects of this disclosure:

[0237] Aspect 1 : An apparatus for wireless communication at a wireless communication device, the apparatus comprising: one or more memories; and one or more processors coupled with the one or more memories, the one or more memories configured to execute code that causes the wireless communication device to: receive an assignment of a set of resources associated with a channel, the set of resources comprising a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal; generate, from the reference signal received on the second subset of resources, a first plurality of channel estimates associated with respective layers of a plurality of layers of the channel for the set of resources according to MMSE operations; generate a second plurality of channel estimates and a plurality of values of a latent variable associated with respective layers of the plurality of layers of the channel for the set of resources according to a nonlinear two-dimensional interpolation of the channel for the set of resources, wherein the nonlinear two-dimensional interpolation of the channel is based at least in part on the first plurality of channel estimates; and perform a refinement operation comprising one or more iterations on the second plurality of channel estimates, wherein each iteration of the one or more iterations is performed according to a same set of machine learning parameters, and wherein, to perform each iteration of the one or more iterations, the one or more processors are configured to cause the wireless communication device to: generate respective gradients associated with the second plurality of channel estimates based at least in part on the second plurality of channel estimates for the second subset of resources and measured observations of the second subset of resources; generate a second set of values of the latent variable based at least in part on a first set of values of the plurality of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradients; and modify the second plurality of channel estimates associated with the plurality of layers based at least in part on the second set of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradients.

[0238] Aspect 2: The apparatus of aspect 14, wherein the refinement operation is a first refinement operation and the set of machine learning parameters is a first set of machine learning parameters, and wherein the one or more processors are configured to cause the wireless communication device to: perform a second refinement operation on the second plurality of channel estimates, the second refinement operation comprising one or more second iterations performed according to a same second set of machine learning parameters, wherein the first refinement operation and the second refinement operation are associated with respective attention computations of a plurality of attention computations.

[0239] Aspect 3: The apparatus of aspect 15, wherein the plurality of attention computations comprises intra-PRB group computations, inter-PRB group computations, cross-MIMO computations, MLP computations, or any combination thereof.

[0240] Aspect 4: The apparatus of any one of aspects 14 through 16, wherein the one or more processors are configured to cause the wireless communication device to: perform the MMSE operation based on a resource configuration mode of the second subset of resources allocated for the reference signal, the reference signal comprising a DMRS.

[0241] Aspect 5: The apparatus of any one of aspects 14 through 17, wherein to generate the respective gradients, the one or more processors are configured to cause the wireless communication device to: generate a respective set of values of a residual variable based at least in part on a difference between the measured observation of the second subset of resources and the second plurality of channel estimates for the second subset of resources; and combine the respective sets of values of the residual variable, the second subset of resources, and a number of mask bits.

[0242] Aspect 6: The apparatus of any one of aspects 14 through 18, wherein to generate the second set of values of the latent variable, the one or more processors are configured to cause the wireless communication device to: combine, based at least in part on generating the respective gradients, the second plurality of channel estimates for the second subset of resources, the respective gradients, and respective values of the first set of values of the latent variable.

[0243] Aspect 7: The apparatus of any one of aspects 14 through 19, wherein to generate the second set of values of the latent variable, the one or more processors are configured to cause the wireless communication device to: model a correlation between resources of each resource block in a resource block group and other resource blocks in the resource block group.

[0244] Aspect 8: The apparatus of any one of aspects 14 through 19, wherein to generate the second set of values of the latent variable, the one or more processors are configured to cause the wireless communication device to: model a correlation between resources of each group in a plurality of groups of resources of the resource set and other groups in the plurality of groups of resources, wherein each group in the plurality of groups of resources comprises a plurality of resource blocks.

[0245] Aspect 9: The apparatus of any one of aspects 14 through 19, wherein to generate the second set of values of the latent variable, the one or more processors are configured to cause the wireless communication device to: model a correlation between each layer in the plurality of layers of the resource set.

[0246] Aspect 10: The apparatus of any one of aspects 14 through 22, wherein to modify the second plurality of channel estimates, the one or more processors are configured to cause the wireless communication device to combine the second set of values of the latent variables, the second plurality of channel estimates, and the respective gradients based at least in part on the set of machine learning parameters.

[0247] Aspect 11 : The apparatus of any one of aspects 14 through 23, wherein the non-linear two-dimensional interpolation of the channel is based at least in part on a machine learning model.

[0248] Aspect 12: The apparatus of any one of aspects 14 through 24, wherein the first plurality of channel estimates and the second plurality of channel estimates are associated with a plurality of SISO antenna pairs.

[0249] Aspect 13: The apparatus of any one of aspects 14 through 25, wherein each iteration of the one or more iterations is performed by a refinement network comprising a likelihood module, an encoder module, and a decoder module, the refinement network comprising a machine learning model; and each refinement network is performed according to the same set of machine learning parameters.

[0250] Aspect 14: A method for wireless communication at a wireless communication device, comprising: receiving an assignment of a set of resources associated with a channel, the set of resources comprising a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal; generating, from the reference signal received on the second subset of resources, a first plurality of channel estimates associated with respective layers of a plurality of layers of the channel for the set of resources according to a MMSE operation; generating a second plurality of channel estimates and a plurality of values of a latent variable associated with respective layers of the plurality of layers of the channel for the set of resources according to a nonlinear two-dimensional interpolation of the channel for the set of resources, wherein the nonlinear two-dimensional interpolation of the channel is based at least in part on the first plurality of channel estimates; and performing a refinement operation comprising one or more iterations on the second plurality of channel estimates, wherein each iteration of the one or more iterations is performed according to a same set of machine learning parameters, and wherein each iteration of the one or more iterations comprises: generating a respective gradient associated with the second plurality of channel estimates based at least in part on the second plurality of channel estimates for the second subset of resources and a measured observation of the second subset of resources; generating a second set of values of the latent variable based at least in part on a first set of values of the plurality of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradient; and modifying the second plurality of channel estimates associated with the plurality of layers based at least in part on the second set of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradient.

[0251] Aspect 15: The method of aspect 14, wherein the refinement operation is a first refinement operation and the set of machine learning parameters is a first set of machine learning parameters, the method further comprising: performing a second refinement operation on the second plurality of channel estimates, the second refinement operation comprising one or more second iterations performed according to a same second set of machine learning parameters, wherein the first refinement operation and the second refinement operation are associated with respective attention computations of a plurality of attention computations.

[0252] Aspect 16: The method of aspect 15, wherein the plurality of attention computations comprises intra-PRB group computations, inter-PRB group computations, cross-MIMO computations, MLP computations, or any combination thereof.

[0253] Aspect 17: The method of any of aspects 14 through 16, further comprising: performing the MMSE operation based on a resource configuration pattern of the second subset of resources allocated for the reference signal, the reference signal comprising a DMRS.

[0254] Aspect 18: The method of any one of aspects 14 through 17, wherein generating the respective gradients comprises: generating a respective set of values of a residual variable based at least in part on a difference between the measured observations of the second subset of resources and the second plurality of channel estimates for the second subset of resources; and combining the respective sets of values of the residual variable, the second subset of resources, and a number of mask bits.

[0255] Aspect 19: The method of any one of aspects 14 through 18, wherein generating the second set of values of the latent variable comprises: combining the second plurality of channel estimates for the second subset of resources, the respective gradients, and respective values of the first set of values of the latent variable based at least in part on generating the respective gradients.

[0256] Aspect 20: The method of any one of aspects 14 through 19, wherein generating the second set of values of the latent variable comprises: modeling a correlation between resources of each resource block in a resource block group and other resource blocks in the resource block group.

[0257] Aspect 21: The method of any one of aspects 14 through 19, wherein generating the second set of values of the latent variable comprises: modeling a correlation between resources of each group in a plurality of groups of resources of the resource set and other groups in the plurality of groups of resources, wherein each group in the plurality of groups of resources comprises a plurality of resource blocks.

[0258] Aspect 22: The method of any one of aspects 14 through 19, wherein generating the second set of values of the latent variable comprises: modeling a correlation between each layer in the plurality of layers of the resource set.

[0259] Aspect 23: The method of any one of aspects 14 through 22, wherein modifying the second plurality of channel estimates comprises: combining the second set of values of the latent variable, the second plurality of channel estimates, and the respective gradients based at least in part on the set of machine learning parameters.

[0260] Aspect 24: The method of any one of aspects 14 through 23, wherein the non-linear two-dimensional interpolation of the channel is based at least in part on a machine learning model.

[0261] Aspect 25: The method of any one of aspects 14 through 24, wherein the first plurality of channel estimates and the second plurality of channel estimates are associated with a plurality of SISO antenna pairs.

[0262] Aspect 26: The method of any one of aspects 14 through 25, wherein each iteration of the one or more iterations is performed by a refinement network comprising a likelihood module, an encoder module, and a decoder module, the refinement network comprising a machine learning model; and each refinement network is performed according to the same set of machine learning parameters.

[0263] Aspect 27: An apparatus for wireless communication at a wireless communication device, the apparatus comprising at least one means for performing a method of any one of aspects 14 through 26.

[0264] Aspect 28: A non-transitory computer-readable medium storing code for wireless communication at a wireless communication device, the code comprising instructions executable by one or more processors to cause the wireless communication device to perform a method of any one of aspects 14 through 26.

[0265] Aspect 29: A method for wireless communication, the method comprising: receiving an assignment of a set of resources associated with a channel, the set of resources comprising a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal; generating a plurality of channel estimates associated with respective layers of a plurality of layers of the channel for the set of resources; and performing a refinement operation comprising one or more iterations on the plurality of channel estimates, wherein each iteration of the one or more iterations comprises: generating respective gradients associated with the plurality of channel estimates based at least in part on the plurality of channel estimates for the second subset of resources and a measured observation of the second subset of resources; generating a second set of values of a latent variable based at least in part on a first set of values of the latent variable, the plurality of channel estimates, and the respective gradients; and modifying the plurality of channel estimates associated with the plurality of layers based at least in part on the second set of values of the latent variable, the plurality of channel estimates, and the respective gradients.

[0266] Aspect 30: The method of aspect 14, wherein generating the respective gradients comprises: generating a respective set of values of a residual variable based at least in part on a difference between the measured observation of the second subset of resources and the plurality of channel estimates for the second subset of resources; and combining the respective set of values of the residual variable, the measured observation of the second subset of resources, and a number of mask bits.

[0267] Aspect 31: The method of any one of aspects 14 through 15, wherein generating the second set of values of the latent variable comprises: combining, based at least in part on generating the respective gradients, the plurality of channel estimates for the second subset of resources, the respective gradients, and respective values of the first set of values of the latent variable.

[0268] Aspect 32: The method of any one of aspects 14-16, wherein generating the second set of values of the latent variable comprises modeling a correlation between resources of each resource block of a resource block group and other resource blocks of the resource block group.

[0269] Aspect 33: The method of any one of aspects 14-17, wherein generating the second set of values of the latent variable comprises modeling a correlation between resources of each group of a plurality of groups of resources of the resource set and other groups of the plurality of groups of resources, wherein each group of the plurality of groups of resources comprises a plurality of resource blocks.

[0270] Aspect 34: The method of any one of aspects 14-18, wherein generating the second set of values of the latent variable comprises modeling a correlation between each layer of the plurality of layers of the resource set.

[0271] Aspect 35: The method of any one of aspects 14-19, wherein modifying the plurality of channel estimates comprises combining the second set of values of the latent variable, the plurality of channel estimates, and the respective gradients.

[0272] Aspect 36: The method of any one of aspects 14-20, wherein initial values of the plurality of channel estimates are associated with SISO antenna pairs.

[0273] Aspect 37: The method of any one of aspects 14-21, wherein the second subset of resources is configured according to a resource configuration mode of a set of resource configuration modes.

[0274] Aspect 38: The method of aspect 22, wherein the set of resource configuration modes is a set of demodulation reference signal modes.

[0275] Aspect 39: The method of any one of aspects 14-23, wherein each iteration is performed by a refinement network comprising a likelihood module, an encoder module, and a decoder module, and each refinement network further comprises respective parameters associated with machine learning operations.

[0276] Aspect 40: The method of any one of aspects 14-24, wherein the resource set comprises one or more groups of resources, and each respective layer of the plurality of layers is associated with a respective antenna pair of a plurality of SISO antenna pairs.

[0277] Aspect 41 : An apparatus for wireless communication, comprising: one or more memories storing processor-executable code; and one or more processors coupled with the one or more memories and capable of individually or collectively executing the code to cause the apparatus to perform the method of any of aspects 14 through 25.

[0278] Aspect 42 : An apparatus for wireless communication, comprising at least one means for performing the method of any of aspects 14 through 25.

[0279] Aspect 43 : A non-transitory computer-readable medium storing code for wireless communication, the code comprising instructions executable by a processor to perform the method of any of aspects 14 through 25.

[0280] It should be noted that the methods described herein describe possible implementations, and that the operations and the steps can be rearranged or otherwise modified and that other implementations are possible. Further, aspects from two or more of the methods can be combined.

[0281] Although aspects of an LTE, LTE-A, LTE-A Pro, or NR system can be described for purposes of example, and LTE, LTE-A, LTE-A Pro, or NR terminology can be used in much of the description, the techniques described herein are applicable beyond LTE, LTE-A, LTE-A Pro, or NR networks. For example, the described techniques can be applicable to various other wireless communication systems, such as Ultra Mobile Broadband (UMB), Institute of Electrical and Electronics Engineers (IEEE) 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20, Flash-OFDM, and others.

[0282] Information and signals described herein can be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips that can be referenced throughout the description can be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof.

[0283] The various illustrative blocks and components described herein can be implemented using at least one general purpose processor (or processing circuitry), a DSP, an ASIC, a CPU, a FPGA, or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor can be a microprocessor, but in the alternative, the processor can be any processor, controller, microcontroller, or state machine. A processor can also be implemented as a combination of computing devices (e.g., a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration). The various functions and operations described herein can be performed by one or more processors as collectively, or individually, capable of performing the functions and operations described herein.

[0284] The functions described herein can be implemented in hardware, software executed by a processor, firmware, or any combination thereof. If implemented in software executed by a processor, the functions can be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Other examples and implementations are within the scope of the disclosure and appended claims. For example, due to the nature of software, functions described herein can be implemented using software executed by a processor, hardware, firmware, hardwiring, or combinations of any of these. Features implementing functions can also be physically located at various positions, including being distributed such that portions of functions are implemented at different physical locations.

[0285] Computer-readable media includes both non-transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium can be any available medium that can be accessed by a general purpose or special purpose computer. By way of example, and not limitation, non-transitory computer-readable media can include RAM, ROM, electrically erasable programmable ROM (EEPROM), flash memory, compact disk (CD) ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to carry or store desired program code means in the form of instructions or data structures and that can be accessed by a general-purpose or special-purpose computer, or a general-purpose or special-purpose processor. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of computer-readable medium. Disk and disc, as used herein, include CD, laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media. Any functions or operations described herein that are capable of being performed by a memory can be performed by a memory circuit and / or multiple memories capable of performing the functions or operations individually or collectively.

[0286] As used herein, including in the claims “or” as used in a list of items (for example, a list of items prefaced by a phrase such as “at least one of’ or “one or more of’) indicates an inclusive list such that, for example, a list of at least one of A, B, or C means A or B or C or AB or AC or BC or ABC (i.e., A and B and C). Also, as used herein, the phrase “based on” shall not be construed as a reference to a closed set of conditions. For example, an example step that is described as “based on condition A” can be based on both a condition A and a condition B without departing from the scope of the present disclosure. In other words, as used herein, the phrase “based on” shall be construed in the same manner as the phrase “based at least in part on.”

[0287] As used herein, including in the claims, the article “a” preceding a noun is open, and is understood to refer to “at least one” or “one or more” of the noun. Thus, the terms “a,” “at least one,” “one or more,” and “at least one of one or more” are interchangeable. For example, where a claim recites “a” “component” that performs one or more functions, it is understood that each function of the single component, or alternatively, of multiple components performing the same function, can be performed respectively by a single component or any combination of the multiple components. Thus, the term “component” having a particular, named property or performing a particular function can refer to “at least one of one or more components” having the particular, named property or performing the particular function. A subsequent reference to “the component” in the claim can thus be understood to refer to any or all of the one or more components. For example, a component introduced with the article “a” can be understood as meaning “one or more components,” and a subsequent reference in the claim to “the component” can be understood as meaning “at least one of the one or more components.” Similarly, a subsequent reference to “the component” introduced with the article “the” or “said” can refer to any or all of the one or more components. For example, a subsequent reference in the claim to “the one or more components” can be understood as meaning “at least one of the one or more components.” As used herein, including in the claims, the terms “set” or “subset” can be understood to refer to one or more items. For example, a reference to “a set of objects” or “a subset of objects” can be understood as equivalently referring to one object or multiple objects.

[0288] The term “determining” encompasses a wide variety of actions and, therefore, “determining” can include calculating, computing, processing, deriving, investigating, looking up (such as via a table, a database or another data structure), ascertaining and the like. Also, “determining” can include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory) and the like. Also, “determining” can include resolving, selecting, choosing, establishing and other such similar actions.

[0289] In the drawings, like components or features can have the same reference label. Also, various components of the same type can be distinguished by adding a dash and a second label that distinguishes among the components of like type. If only the first reference label is used in the specification, the description is applicable to any one of the components having the same first reference label irrespective of the second reference label, or other subsequent reference label.

[0290] The description set forth herein, in connection with the appended drawings, describes example configurations and does not represent all the examples that can be implemented or that are within the scope of the claims. The term “example” used herein means “serving as an example, instance, or illustration,” and not “preferred” over other examples. The detailed description includes specific details for the purpose of providing an understanding of the described techniques. These techniques, however, can be practiced without these specific details. In some instances, well-known structures and devices are shown in block diagram form in order to avoid obscuring the concepts of the described examples.

[0291] The description herein is presented to enable any person skilled in the art to practice the present disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other variations without departing from the scope of the disclosure. Thus, the present disclosure is not to be limited to the examples and designs described herein but is to be accorded the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. An apparatus for wireless communication at a wireless communication device, the apparatus comprising: one or more memories; and one or more processors coupled with the one or more memories and configured to cause the wireless communication device to: receive an assignment of a set of resources associated with a channel, the set of resources comprising a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal; generate, from the reference signal received on the second subset of resources, a first plurality of channel estimates associated with respective layers of a plurality of layers of the channel for the set of resources according to a least squares estimation operation; generate a second plurality of channel estimates and a plurality of values of a latent variable associated with respective layers of the plurality of layers of the channel for the set of resources according to a nonlinear two-dimensional interpolation of the channel for the set of resources, wherein the nonlinear two-dimensional interpolation of the channel is based at least in part on the first plurality of channel estimates; and perform, on the second plurality of channel estimates, a refinement operation comprising one or more iterations, wherein each iteration of the one or more iterations is performed according to a same set of machine learning parameters, and wherein, to perform each iteration of the one or more iterations, the one or more processors are configured to cause the wireless communication device to: generate respective gradients associated with the second plurality of channel estimates based at least in part on the second plurality of channel estimates for the second subset of resources and measured observations of the second subset of resources; generate a second set of values of the latent variable based at least in part on a first set of values of the plurality of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradients; and modify the second plurality of channel estimates associated with the plurality of layers based at least in part on the second set of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradients.

2. The apparatus of claim 1, wherein the refinement operation is a first refinement operation and the set of machine learning parameters is a first set of machine learning parameters, and the one or more processors are configured to cause the wireless communication device to: perform a second refinement operation on the second plurality of channel estimates, the second refinement operation comprising one or more second iterations performed according to a same second set of machine learning parameters, wherein the first refinement operation and the second refinement operation are associated with respective attention computations of a plurality of attention computations.

3. The apparatus of claim 2, wherein the plurality of attention computations comprises intra-physical resource block group computations, inter-physical resource block group computations, cross multiple-input multiple-output computations, multi-layer perceptron computations, or any combination thereof.

4. The apparatus of claim 1, wherein the one or more processors are configured to cause the wireless communication device to: performing the minimum mean square estimation operation based on a resource configuration pattern of the second subset of resources allocated for the reference signal, the reference signal comprising a demodulation reference signal.

5. The apparatus of claim 1, wherein, To generate the respective gradient, the one or more processors are configured to cause the wireless communication device to: generate a respective set of values of a residual variable based at least in part on a difference between the measured observation of the second subset of resources and the second plurality of channel estimates for the second subset of resources; and combine the respective set of values of the residual variable, the second subset of resources, and a number of mask bits.

6. The apparatus of claim 1, wherein, To generate the second set of values of the latent variable, the one or more processors are configured to cause the wireless communication device to: combine the second plurality of channel estimates for the second subset of resources, the respective gradient, and a respective value of the first set of values of the latent variable based at least in part on the generation of the respective gradient.

7. The apparatus of claim 1, wherein, To generate the second set of values of the latent variable, the one or more processors are configured to cause the wireless communication device to: model a correlation between resources of each resource block in a resource block group and other resource blocks in the resource block group.

8. The apparatus of claim 1, wherein, To generate the second set of values of the latent variable, the one or more processors are configured to cause the wireless communication device to: model a correlation between resources of each group of a plurality of groups of resources of the resource set and other groups in the plurality of groups of resources, wherein each group of the plurality of groups of resources comprises a plurality of resource blocks.

9. The apparatus of claim 1, wherein, To generate the second set of values of the latent variable, the one or more processors are configured to cause the wireless communication device to: model a correlation between each layer of the plurality of layers of the resource set.

10. The apparatus of claim 1, wherein, To modify the second plurality of channel estimates, the one or more processors are configured to cause the wireless communication device to: combine the second set of values of the latent variable, the second plurality of channel estimates, and the respective gradient based at least in part on the set of machine learning parameters.

11. The apparatus of claim 1, wherein the non-linear two-dimensional interpolation of the channel is based at least in part on a machine learning model.

12. The apparatus of claim 1, wherein the first plurality of channel estimates and the second plurality of channel estimates are associated with a plurality of single-input and single-output antenna pairs.

13. The apparatus of claim 1, wherein: each iteration of the one or more iterations is performed by a refinement network comprising a likelihood module, an encoder module, and a decoder module, the refinement network comprising a machine learning model; and each refinement network is performed according to the same set of machine learning parameters.

14. A method for wireless communication at a wireless communication device, the method comprising: receiving an assignment of a resource set associated with a channel, the resource set comprising a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal; generating a first plurality of channel estimates associated with respective ones of a plurality of layers of the channel for the set of resources from the reference signal received on the second subset of resources according to a least squares estimation operation; generating a second plurality of channel estimates and a plurality of values of a latent variable associated with respective ones of the plurality of layers of the channel for the set of resources according to a non-linear two-dimensional interpolation of the channel, wherein the non-linear two-dimensional interpolation of the channel is based at least in part on the first plurality of channel estimates; and performing a refinement operation comprising one or more iterations on the second plurality of channel estimates, wherein each of the one or more iterations is performed according to a same set of machine learning parameters, and wherein each of the one or more iterations comprises: generating a respective gradient associated with the second plurality of channel estimates based at least in part on the second plurality of channel estimates for the second subset of resources and observations of measurements of the second subset of resources; generating a second set of values of the latent variable based at least in part on a first set of values of the plurality of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradient; and modifying the second plurality of channel estimates associated with the plurality of layers based at least in part on the second set of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradient.

15. The method of claim 14, wherein the refinement operation is a first refinement operation and the set of machine learning parameters is a first set of machine learning parameters, the method further comprising: performing a second refinement operation on the second plurality of channel estimates, the second refinement operation comprising one or more second iterations performed according to a same second set of machine learning parameters, wherein the first refinement operation and the second refinement operation are associated with respective ones of a plurality of attention computations.

16. The method of claim 15, wherein the plurality of attention computations comprises intra-physical resource block group computations, inter-physical resource block group computations, cross multiple-input multiple-output computations, multi-layer perceptron computations, or any combination thereof.

17. The method of claim 14, the method further comprising: performing the least squares estimation operation based on a resource configuration pattern of the second subset of resources allocated for the reference signal, the reference signal comprising a demodulation reference signal.

18. The method of claim 14, wherein generating the respective gradient comprises: generating a respective set of values of a residual variable based at least in part on a difference between the observations of the measurements of the second subset of resources and the second plurality of channel estimates for the second subset of resources; and combining the respective set of values of the residual variable, the second subset of resources, and a number of mask bits.

19. The method of claim 14, wherein generating the second set of values of the latent variable comprises: combining the second plurality of channel estimates for the second subset of resources, the respective gradient, and a respective value of the first set of values of the latent variable based at least in part on generating the respective gradient.

20. The method of claim 14, wherein generating the second set of values of the latent variable comprises: modeling a correlation between resources of each resource block of a resource block group and other resource blocks of the resource block group.

21. The method of claim 14, wherein generating the second set of values of the latent variable comprises: modeling a correlation between resources of each group of a plurality of groups of resources of the resource set and other groups of the plurality of groups of resources, wherein each group of the plurality of groups of resources comprises a plurality of resource blocks.

22. The method of claim 14, wherein generating the second set of values of the latent variable comprises: modeling a correlation between each layer of the plurality of layers of the resource set.

23. The method of claim 14, wherein modifying the second plurality of channel estimates comprises: combining the second set of values of the latent variable, the second plurality of channel estimates, and the respective gradient based at least in part on the set of machine learning parameters.

24. The method of claim 14, wherein the non-linear two-dimensional interpolation of the channel is based at least in part on a machine learning model.

25. The method of claim 14, wherein the first plurality of channel estimates and the second plurality of channel estimates are associated with a plurality of single-input and single-output antenna pairs.

26. The method of claim 14, wherein: each iteration of the one or more iterations is performed by a refinement network comprising a likelihood module, an encoder module, and a decoder module, the refinement network comprising a machine learning model; and each refinement network is performed according to the same set of machine learning parameters.

27. A wireless communication device for wireless communication, the wireless communication device comprising: means for receiving an assignment of a resource set associated with a channel, the resource set comprising a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal; means for generating, according to a minimum mean square estimation operation, a first plurality of channel estimates associated with respective layers of a plurality of layers of the channel for the resource set from the reference signal received on the second subset of resources; means for generating, according to a non-linear two-dimensional interpolation of the channel for the resource set, a second plurality of channel estimates and a plurality of values of a latent variable, the second plurality of channel estimates and the plurality of values being associated with respective layers of the plurality of layers of the channel for the resource set, wherein the non-linear two-dimensional interpolation of the channel is based at least in part on the first plurality of channel estimates; and and means for performing a refinement operation comprising one or more iterations on the second plurality of channel estimates, wherein each iteration of the one or more iterations is performed according to a same set of machine learning parameters, and wherein the means for performing each iteration of the one or more iterations comprises: means for generating respective gradients associated with the second plurality of channel estimates based at least in part on the second plurality of channel estimates for the second subset of resources and observations of measurements of the second subset of resources; means for generating a second set of values of the latent variable based at least in part on a first set of values of the plurality of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradients; and means for modifying the second plurality of channel estimates associated with the plurality of layers based at least in part on the second set of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradients.

28. The wireless communication device of claim 27, wherein the refinement operation is a first refinement operation and the set of machine learning parameters is a first set of machine learning parameters, the wireless communication device further comprising: means for performing a second refinement operation on the second plurality of channel estimates, the second refinement operation comprising one or more second iterations performed according to a same second set of machine learning parameters, wherein the first refinement operation and the second refinement operation are associated with respective attention computations of a plurality of attention computations.

29. A non-transitory computer-readable medium storing code for wireless communication at a wireless communication device, the code comprising instructions executable by one or more processors to cause the wireless communication device to: receive an assignment of a set of resources associated with a channel, the set of resources comprising a first subset of resources allocated for a data signal and a second subset of resources allocated for a reference signal; generate, according to a minimum mean square estimation operation, a first plurality of channel estimates associated with respective layers of a plurality of layers of the channel for the set of resources from the reference signal received on the second subset of resources; generate, according to a nonlinear two-dimensional interpolation of the channel for the set of resources, a second plurality of channel estimates and a plurality of values of a latent variable, the second plurality of channel estimates and the plurality of values being associated with respective layers of the plurality of layers of the channel for the set of resources, wherein the nonlinear two-dimensional interpolation of the channel is based at least in part on the first plurality of channel estimates; and perform a refinement operation comprising one or more iterations on the second plurality of channel estimates, wherein each iteration of the one or more iterations is performed according to a same set of machine learning parameters, and wherein the instructions to perform each iteration of the one or more iterations are executable to: generate respective gradients associated with the second plurality of channel estimates based at least in part on the second plurality of channel estimates for the second subset of resources and observations of measurements of the second subset of resources; generating a second set of values of the latent variable based at least in part on a first set of values of the plurality of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradients; and modifying the second plurality of channel estimates associated with the plurality of layers based at least in part on the second set of values of the latent variable, the second plurality of channel estimates, the set of machine learning parameters, and the respective gradients.

30. The non-transitory computer-readable medium of claim 29, wherein the refinement operation is a first refinement operation and the set of machine learning parameters is a first set of machine learning parameters, and the instructions are further executable by the one or more processors to cause the wireless communication device to: perform a second refinement operation on the second plurality of channel estimates, the second refinement operation comprising one or more second iterations performed according to a same second set of machine learning parameters, wherein the first refinement operation and the second refinement operation are associated with respective attention computations of a plurality of attention computations.

Images