Method and apparatus for denoising a noisy representation of a wireless channel

Abstract

An embodiment disclosed herein is a method for performing channel estimation. The method includes obtaining a beam-delay representation of the wireless channel in a beam-delay domain in which channel energy is expected to be concentrated on delay taps located in a finite portion of a delay tap dimension, obtaining a positional encoding that encodes position information of beams and delay taps in the beam-delay domain using a waveform that increases in frequency along the delay tap dimension when going in a direction away from the finite portion of the delay tap dimension, superimposing the positional encoding on the beam-delay representation of the wireless channel to generate an augmented beam-delay representation of the wireless channel, and applying a machine learning model to the augmented beam-delay representation of the wireless channel to generate a denoised beam-delay representation of the wireless channel.

Description

METHOD AND APPARATUS FOR DENOISING A NOISY REPRESENTATION OF AWIRELESS CHANNELTECHNICAL FIELD

[0001] Embodiments of the invention relate to the field of wireless communication systems, and more specifically to denoising a noisy representation of a wireless channel.BACKGROUND

[0002] Channel estimation refers to the process of estimating the characteristics of a wireless channel such as scattering, fading, and power decay with distance. Channel estimation allows the transmitter and / or receiver of a wireless communication to adapt to the channel conditions, which is important for achieving reliable communication with high data rates in multi-antenna systems. In Fifth Generation (5G) New Radio (NR), channel estimation is performed based on a transmitter transmitting various known reference signals such as a channel state information reference signal (CSI-RS, which is used for channel sounding in downlink (DL)), a sounding reference signal (SRS, which is used for channel sounding in uplink (UL)), and / or a demodulation reference signal (DMRS, which is used for data demodulation in DL and UL) to the receiver. The receiver may estimate the characteristics of the channel based on comparing the actual signal received at the receiver to the known reference signal. However, it is difficult to accurately estimate the characteristics of the channel due to the presence of noise and interference.

[0003] During channel estimation, an initial representation of the channel (e.g., a mathematical model of the channel characteristics) may be generated based on the received reference signal. The initial representation of the channel may be noisy due to the aforementioned presence of noise and interference. A denoising technique may be applied to the noisy representation of the channel to remove noise from the noisy representation of the channel. Conventional denoising techniques attempt to remove noise from the noisy representation of the channel based on analyzing the statistical properties of the channel. However, conventional denoising techniques are limited in the amount of noise they can remove from the noisy representation of the channel.SUMMARY

[0004] Disclosed herein is a method performed by a computing device to denoise a representation of a wireless channel between a first wireless communication device and a second wireless communication device. The method includes obtaining a beam-delay representation of the wireless channel in a beam-delay domain in which channel energy is expected to beconcentrated on delay taps located in a finite portion of a delay tap dimension, obtaining a positional encoding that encodes position information of beams and delay taps in the beam-delay domain using a waveform that increases in frequency along the delay tap dimension when going in a direction away from the finite portion of the delay tap dimension, superimposing the positional encoding on the beam-delay representation of the wireless channel to generate an augmented beam-delay representation of the wireless channel, and applying a machine learning model to the augmented beam-delay representation of the wireless channel to generate a denoised beam-delay representation of the wireless channel.

[0005] Disclosed herein is a computing device that is configured to denoise a representation of a wireless channel between a first wireless communication device and a second wireless communication device. The computing device includes a set of one or more processors and a non-transitory machine-readable storage medium that provides instructions that, if executed by the set of one or more processors will cause the computing device to carry out operations for performing channel estimation for a wireless channel between a first wireless communication device and a second wireless communication device. The operations include obtaining a beamdelay representation of the wireless channel in a beam-delay domain in which channel energy is expected to be concentrated on delay taps located in a finite portion of a delay tap dimension, obtaining a positional encoding that encodes position information of beams and delay taps in the beam-delay domain using a waveform that increases in frequency along the delay tap dimension when going in a direction away from the finite portion of the delay tap dimension, superimposing the positional encoding on the beam-delay representation of the wireless channel to generate an augmented beam-delay representation of the wireless channel, and applying a machine learning model to the augmented beam-delay representation of the wireless channel to generate a denoised beam-delay representation of the wireless channel.

[0006] Disclosed herein is a non-transitory machine-readable storage medium that provides instructions that, if executed by a processor of a computing device, will cause the computing device to carry out operations for denoising a representation of a wireless channel between a first wireless communication device and a second wireless communication device. The operations include obtaining a beam-delay representation of the wireless channel in a beamdelay domain in which channel energy is expected to be concentrated on delay taps located in a finite portion of a delay tap dimension, obtaining a positional encoding that encodes position information of beams and delay taps in the beam-delay domain using a waveform that increases in frequency along the delay tap dimension when going in a direction away from the finite portion of the delay tap dimension, superimposing the positional encoding on the beam-delay representation of the wireless channel to generate an augmented beam-delay representation ofthe wireless channel, and applying a machine learning model to the augmented beam-delay representation of the wireless channel to generate a denoised beam-delay representation of the wireless channel.

[0007] An advantage of the disclosed techniques is that, with the disclosed techniques, a multiantenna system is able to remove more noise from the noisy representation of a channel compared to prior techniques. In particular, by encoding most of the position information in the sparse region of the channel in the beam-delay domain, position information can be provided to a ML model, while minimally interfering with the underlying channel pattern. As such, the ML model is able to better learn the underlying pattern of the wireless channel, which in turn helps the ML model to better denoise the beam-delay representation of the wireless channel (i.e., the ML model can better recover the true channel characteristics). Accordingly, by using the disclosed techniques, the multi -antenna system is able to achieve more reliable communication with higher data rates relative to conventional techniques. Also, more generally, the disclosed techniques can be used to assist ML models that take a representation of the wireless channel as input with leaming / inferencing.BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The invention may best be understood by referring to the following description and accompanying drawings that are used to illustrate particular embodiments of the invention. In the drawings:

[0009] Figure l is a diagram showing components of channel estimation system, according to some embodiments.

[0010] Figure 2 is a diagram showing further details of the noise filtering component, according to some embodiments.

[0011] Figure 3 is a diagram showing graph representations of the real and imaginary parts of a PE, according to some embodiments.

[0012] Figure 4 is a diagram showing graph representations of the real and imaginary parts of a representation of a channel in the beam-delay domain, according to some embodiments.

[0013] Figure 5 is a diagram showing graph representations of the real and imaginary parts of an augmented representation of the channel, according to some embodiments.

[0014] Figure 6 is a diagram showing graph representations of the real and imaginary parts of a representation of the channel that is superimposed with a conventional PE, according to some embodiments.

[0015] Figure 7 is a flow diagram of a method for denoising a representation of a wireless channel, according to some embodiments.

[0016] Figure 8 is a diagram showing an environment in which a representation of a channel can be denoised, according to some embodiments.

[0017] Figure 9 is a diagram showing three examples of a network device that can denoise a representation of a channel, according to some embodiments.DETAILED DESCRIPTION

[0018] The following description describes methods and apparatus for denoising a noisy representation of a wireless channel. In the following description, numerous specific details such as logic implementations, opcodes, means to specify operands, resource partitioning / sharing / duplication implementations, types and interrelationships of system components, and logic partitioning / integration choices are set forth in order to provide a more thorough understanding of the present invention. It will be appreciated, however, by one skilled in the art that the invention may be practiced without such specific details. In other instances, control structures, gate level circuits and full software instruction sequences have not been shown in detail in order not to obscure the invention. Those of ordinary skill in the art, with the included descriptions, will be able to implement appropriate functionality without undue experimentation.

[0019] References in the specification to “one embodiment,” “an embodiment,” “an example embodiment,” etc., indicate that the embodiment described may include a particular feature, structure, or characteristic, but every embodiment may not necessarily include the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of one skilled in the art to affect such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described.

[0020] Bracketed text and blocks with dashed borders (e.g., large dashes, small dashes, dotdash, and dots) may be used herein to illustrate optional operations that add additional features to embodiments of the invention. However, such notation should not be taken to mean that these are the only options or optional operations, and / or that blocks with solid borders are not optional in certain embodiments of the invention.

[0021] In the following description and claims, the terms “coupled” and “connected,” along with their derivatives, may be used. It should be understood that these terms are not intended as synonyms for each other. “Coupled” is used to indicate that two or more elements, which may or may not be in direct physical or electrical contact with each other, co-operate or interact witheach other. “Connected” is used to indicate the establishment of communication between two or more elements that are coupled with each other.

[0022] An electronic device stores and transmits (internally and / or with other electronic devices over a network) code (which is composed of software instructions and which is sometimes referred to as computer program code or a computer program) and / or data using machine-readable media (also called computer-readable media), such as machine-readable storage media (e.g., magnetic disks, optical disks, solid state drives, read only memory (ROM), flash memory devices, phase change memory) and machine-readable transmission media (also called a carrier) (e.g., electrical, optical, radio, acoustical or other form of propagated signals - such as carrier waves, infrared signals). Thus, an electronic device (e.g., a computer) includes hardware and software, such as a set of one or more processors (e.g., wherein a processor is a microprocessor, controller, microcontroller, central processing unit, digital signal processor, application specific integrated circuit, field programmable gate array, other electronic circuitry, a combination of one or more of the preceding) coupled to one or more machine-readable storage media to store code for execution on the set of processors and / or to store data. For instance, an electronic device may include non-volatile memory containing the code since the non-volatile memory can persist code / data even when the electronic device is turned off (when power is removed), and while the electronic device is turned on that part of the code that is to be executed by the processor(s) of that electronic device is typically copied from the slower nonvolatile memory into volatile memory (e.g., dynamic random access memory (DRAM), static random access memory (SRAM)) of that electronic device. Typical electronic devices also include a set of one or more physical network interface(s) (NI(s)) to establish network connections (to transmit and / or receive code and / or data using propagating signals) with other electronic devices. For example, the set of physical Nis (or the set of physical NI(s) in combination with the set of processors executing code) may perform any formatting, coding, or translating to allow the electronic device to send and receive data whether over a wired and / or a wireless connection. In some embodiments, a physical NI may comprise radio circuitry capable of receiving data from other electronic devices over a wireless connection and / or sending data out to other devices via a wireless connection. This radio circuitry may include transmitted s), received s), and / or transceiver(s) suitable for radiofrequency communication. The radio circuitry may convert digital data into a radio signal having the appropriate parameters (e.g., frequency, timing, channel, bandwidth, etc.). The radio signal may then be transmitted via antennas to the appropriate recipient(s). In some embodiments, the set of physical NI(s) may comprise network interface controlled s) (NICs), also known as a network interface card, network adapter, or local area network (LAN) adapter. The NIC(s) may facilitate in connecting the electronic device toother electronic devices allowing them to communicate via wire through plugging in a cable to a physical port connected to a NIC. One or more parts of an embodiment of the invention may be implemented using different combinations of software, firmware, and / or hardware.

[0023] A network device (ND) is an electronic device that communicatively interconnects other electronic devices on the network (e.g., other network devices, end-user devices). Some network devices are “multiple services network devices” that provide support for multiple networking functions (e.g., routing, bridging, switching, Layer 2 aggregation, session border control, Quality of Service, and / or subscriber management), and / or provide support for multiple application services (e.g., data, voice, and video).

[0024] As mentioned above, it is difficult to accurately estimate the characteristics of a wireless channel (also simply referred to as a “channel” herein) due to the presence of noise and interference. Conventional denoising techniques that are based on analyzing statistical properties of the channel are limited in terms of the amount of noise they can remove from a noisy representation of the channel. Embodiments are disclosed herein that leverage machine learning (ML) and positional encoding (PE) techniques to more effectively remove noise from a noisy representation of a channel in comparison to conventional denoising techniques.

[0025] ML is the study of programs that can improve their performance on a given task automatically. ML techniques train a ML model to perform the given task. Artificial neural networks (ANNs, also shortened to neural networks (NNs) or neural nets) are a branch of ML models that use a design that is inspired by the biological brain. A NN has a collection of connected units or nodes called artificial neurons, which loosely model the neurons in a biological brain.

[0026] Transformers, which are deep feed-forward ANNs with a self-attention mechanism, have been successful in performing various natural language processing (NLP) tasks. PE techniques may help transformers perform better with NLP tasks by providing information about the relative positions of words in a sentence to the ML model. The position and order of words in a sentence are an essential part of any language, as they define the grammar and thus the actual semantics of sentences. Transformers do not have a built-in mechanism to capture the relative positions and order of words in a sentence. PE techniques provide the transformer with some information regarding the positions and order of words in a sentence, and thus allow the transformer to learn the relative positions of words in a sentence. Without the use of PE techniques, the transformer would perceive a sequence of words as a “bag of words” without any sense of positioning / ordering.

[0027] Embodiments leverage ML and PE techniques to denoise a noisy representation of a channel. An embodiment is a method performed by a computing device for denoising arepresentation of a wireless channel between a first wireless communication device and a second wireless communication device. The method includes obtaining a beam-delay representation of the wireless channel in a beam-delay domain in which channel energy is expected to be concentrated on delay taps located in a finite portion of a delay tap dimension, obtaining a PE that encodes position information of beams and delay taps in the beam-delay domain using a waveform that increases in frequency along the delay tap dimension when going in a direction away from the finite portion of the delay tap dimension, superimposing the PE on the beamdelay representation of the wireless channel to generate an augmented beam-delay representation of the wireless channel, and applying a ML model to the augmented beam-delay representation of the wireless channel to generate a denoised beam-delay representation of the wireless channel.

[0028] According to some embodiments, the PE design takes the properties of the channel in the beam-delay domain into consideration. In most practical scenarios, most of the channel is sparse in the beam-delay domain in that most of the channel energy is concentrated on a few beams and delay taps (there are only a few dominant paths that constitute the channel). The PE may be designed such that most of the position information is encoded in the sparse region of the channel in the beam-delay domain, so as not to interfere with the underlying channel pattern when the PE is superimposed on the beam-delay representation of the channel. Thus, embodiments are able to provide position information to the ML model while minimally interfering with the underlying channel pattern, which helps the ML model to better learn the underlying pattern of the channel, which in turn helps the ML model to better denoise the beamdelay representation of the channel (i.e., the ML model can better recover the true channel characteristics). The ability to better remove noise from a representation of the channel may provide the ability to achieve more reliable communication with higher data rates in multiantenna systems.

[0029] Conventional PEs are designed primarily for NLP use cases. Applying the conventional PE design to a noisy representation of a channel in the complex domain may not provide any assistance to the ML model, and may even degrade the denoising performance of the ML model due to the PE interfering with the underlying channel pattern encoded in the representation of the channel. In contrast, the PE design disclosed herein avoids this drawback by encoding most of the position information in the sparse region of the channel in the beamdelay domain (a region of the channel that is expected to have minimal channel information). Example embodiments are further described herein with reference to the accompanying figures.

[0030] Figure l is a diagram showing components of channel estimation system, according to some embodiments.

[0031] As shown in the diagram, the channel estimation system includes a match filtering component 110, a beam space transformation component 115, a delay compensation component 120, a delay domain transformation component 125, a noise filtering component 130, an inverse delay domain transformation component 135, an inverse delay compensation component 140, and an inverse beam space transformation component 145.

[0032] In the following description, channel estimation for a single reference signal (RS) time resource is considered for sake of simplicity. That is, the description below ignores the time index, t. It should be appreciated, however, that the channel estimation technique disclosed herein can be extended to apply to multiple RS time resources.

[0033] As shown in the diagram, the match filtering component 110 obtains a reference signal 105 as input. The reference signal 105 may have been received by a receiving wireless communication device from a wireless transmitting device over a channel. The transmitting wireless communication device may have generated and transmitted the reference signal 105 to the receiving wireless communication device for the purpose of performing channel estimation. The reference signal 105 may contain noise when it is received by the receiving wireless communication device due to the presence of noise (e.g., caused by thermal agitation of a circuit) and / or interference (e.g., caused by undesired / unexpected signals received at the receiving wireless communication device such as inter-cell and / or intra-cell interference). The reference signal 105 may be a predefined signal that is known to the channel estimation system. Thus, the channel estimation system may have pre-knowledge of what a non-noisy reference signal should look like. In an embodiment, the reference signal 105 is a channel state information reference signal (CSI-RS), a sounding reference signal (SRS), or a demodulation reference signal (DMRS). In an embodiment, the receiving wireless communication device is a base station in a mobile network (e.g., a gNB) and the transmitting wireless communication device is a user equipment (UE) in the mobile network (e.g., a smartphone, a tablet, a laptop, or other device that can wirelessly connect to the mobile network) or vice versa. For sake of illustration, the following description assumes a context where a UE is the transmitting wireless communication device (the device that transmits the reference signal) and a gNB is the receiving wireless communication device (the device that receives the reference signal). It should be appreciated that the roles of the UE and gNB can be switched. In an embodiment, one or more components of the channel estimation system are implemented by a wireless communication device. For example, one or more components of the channel estimation system may be implemented by a base station in a mobile network (e.g., a gNB) or a user equipment (UE) in mobile network. In an embodiment, one or more components of the channel estimation system are implemented by a separate computing / network device (e.g., a computing / network device thatis separate from the wireless communication devices transmitting / receiving the reference signal 105).

[0034] The match filtering component 110 may generate an initial representation of the channel based on the reference signal 105. The initial representation of the channel may be a mathematical model (e.g., a matrix) of the channel characteristics. In an embodiment, the match filtering component 110 generates the initial representation of the channel across each RS port of the UE (denoted by Hm[gNB ports, 1]) according to equation 1 :

[0035] (1) Hm= [gNB ports, 1] sf,m

[0036] In equation 1, H -mis the initial (noisy) representation of the channel, m G{0, ••• , M — 1], Mis the number of UE ports / antennas, y^-is the received reference signal at frequency and Symis the transmitted reference signal transmitted at frequency / from UE port m. The bracket notation indicates the dimensions of a variable (e.g., in equation 1, Hmhas a dimension of “gNB ports x 1,” meaning that it has “gNB ports” rows and one column). Based on equation 1, the channel from all of the UE ports to the gNB ports across all of the frequency resources may be defined as H [gNB ports, mRS, UE ports], where gNB ports is the number of gNB ports, mRSis the total number of RS across the frequency resource, and UE ports is the number of UE ports. The initial representation of the channel is in the antennafrequency domain across the RS time-frequency resources. The initial representation of the channel is noisy because it is generated based on a reference signal 105 that is transmitted in the presence of noise and interference. Thus, the initial representation of the channel may be referred to herein as a noisy representation of the channel in the antenna-frequency domain (or as a noisy antenna-frequency representation of the channel). As used herein, the antennafrequency domain refers to a domain in which the channel is represented with respect to antenna indexes and sub-carrier (frequency) indexes. The match filtering component 110 may provide the noisy antenna-frequency representation of the channel to the beam space transformation component 115.

[0037] The beam space transformation component 115, delay compensation component 120, and the delay domain transformation component 125 may perform operations to transform the noisy antenna-frequency representation of the channel from the antenna-frequency domain to the beam-delay domain. Transforming the noisy antenna-frequency representation of the channel to the beam-delay domain is expected to concentrate the channel energy on a few beams and delay taps, thereby localizing the underlying channel pattern to a smaller region. As used herein, the beam-delay domain refers to a domain in which the channel is represented with respect to angle / beam indexes and delay tap indexes. The beam delay domain is a linear transformation ofthe antenna-frequency domain. The beam space transformation component 115, delay compensation component 120, and the delay domain transformation component 125 are described in further detail below.

[0038] The beam space transformation component 115 may perform beam space transformation. In an embodiment, the beam space transformation component 115 transforms the noisy antenna-frequency representation of the channel to the beam domain across the gNB ports according to equation 2:

[0039] (2) H = UH [beams, mRS, UE ports]

[0040] In equation 2, H is the beam space transformation result, U is the beam space transformation, H is the noisy antenna-frequency representation of the channel, beams is the number of beams, mRSis the total number of RS across the frequency resource, and UE ports is the number of UE ports.

[0041] Assuming that the number of beams is the same as the number of gNB antennas, for dual-polarized antennas, the beam space transformation may be defined according to equations 3-6:

[0042] (3) U = kron(M?MHMR) [beams, gNB ports]

[0046] In equation 3-6, kron is the Kronecker tensor product, H and V represent the columns and rows of the gNB port array, respectively. It is assumed that the antenna index follows the row first ordering (i.e., the index(p, v, h) = p * HV + v * H + h, for p = 0,1, v = 0: V — 1, and h = 0: H — 1).

[0047] The beam space transformation component 115 may provide the beam space transformation result to the delay compensation component 120.

[0048] The delay compensation component 120 may apply delay compensation to the beam space transformation result in the frequency domain. In an embodiment, the delay compensation component 120 applies delay compensation according to equation 7:

[0050] In equation 7, H[.y .] is a delay compensation result, = n *A, h. is the index of the pilot (or RS) position in the frequency grid, is the subcarrier spacing, and fAis the delay compensation for the reference symbol (e.g., which is applied to compensate for different arrival times of orthogonal frequency division multiplexing (OFDM) symbols). This induces H [beams, mRS, UE ports], where beams is the number of beams, mRSis the total number of RS across the frequency resource, and UE ports is the number of UE ports.

[0051] The delay compensation component 120 may provide the delay compensation result to the delay domain transformation component 125.

[0052] The delay domain transformation component 125 may perform delay domain transformation. In an embodiment, the delay domain transformation component 125 applies a discrete cosine transform (DCT) to the delay compensation result (e.g., H [beams,:, UE ports]) to generate a noisy beam-delay representation of the channel according to equation 8:

[0053] (8) H = DCT(H), [beams, taps, UE ports]

[0054] In equation 8, H is the noisy beam-delay representation of the channel, DCT is a discrete cosine transform, H is the delay compensation result, beams is the number of beams, and UE ports is the number of UE ports.

[0055] The delay domain transformation component 125 may provide the noisy beam-delay representation of the channel to the noise filtering component 130.

[0056] The noise filtering component 130 may denoise the noisy beam-delay representation of the channel. As will be described in further detail herein, the noise filtering component 130 may leverage ML and PE techniques to denoise the noisy beam-delay representation of the channel. The noise filtering component 130 may generate a denoised beam-delay representation of the channel (e.g., H [beams, taps, UE ports]) and provide it to the inverse delay domain transformation component 135.

[0057] The inverse delay domain transformation component 135, the inverse delay compensation component 140, and the inverse beam space transformation component 145 may perform operations to transform the denoised beam-delay representation of the channel from the beam-delay domain to the antenna-frequency domain. The inverse delay domain transformation component 135, the inverse delay compensation component 140, and the inverse beam space transformation component 145 are described in further detail below.

[0058] The inverse delay domain transformation component 135 may perform an inverse delay domain transformation. In an embodiment, the inverse delay domain transformation component 135 applies an inverse DCT (IDCT) to the denoised beam-delay representation of thechannel (H [beams, UE ports]) along the delay tap dimension after zero padding to interpolate across the entire bandwidth, for example, according to equation 9:

[0059] (9) H = / DCT(H), [beams, kc* mRS, UE ports]

[0060] In equation 9, H is the inverse delay domain transformation result, IDCT is an inverse DCT, H is the denoised beam-delay representation of the channel, beams is the number of beams, kcis the RS comb used to spread the RS across the bandwidth such that every kcfrequency resource is occupied by the RS, mRSis the total number of RS across the frequency resource, and UE ports is the number of ports.

[0061] The inverse delay domain transformation component 135 may provide the inverse delay domain transformation result to the inverse delay compensation component 140.

[0062] The inverse delay compensation component 140 may apply inverse delay compensation to the inverse delay domain transformation result in the frequency domain. In an embodiment, the inverse delay compensation component 140 applies inverse delay compensation according to equation 10:

[0064] In equation 10, H[.y .] is the inverse delay compensation result, H[.y .] is the inverse delay domain transformation result, f = n * f^ n is the index of the pilot (or RS) position in the frequency grid across the entire bandwidth,is the subcarrier spacing, and tAis the delay compensation for the reference symbol (e.g., which is applied to compensate for different arrival times of orthogonal frequency division multiplexing (OFDM) symbols). This induces H [beams, kc* mRS, UE ports], where beams is the number of beams, kcis the RS comb used to spread the RS across the bandwidth such that every kcfrequency resource is occupied by the RS, mRSis the total number of RS across the frequency resource, and UE ports is the number of UE ports.

[0065] The inverse beam space transformation component 145 may perform inverse beam space transformation. In an embodiment, the inverse beam space transformation component 145 transforms the inverse delay compensation result to the antenna domain according to equation 11 :

[0066] (11) H = UHH [gNB ports, kc* mRS, UE ports]

[0067] In equation 11, H is the denoised antenna-frequency representation of the channel, UHis a hermitian transpose of the beam space transformation, H is the inverse delay compensation result, gNB ports is the number of gNB ports, kcis the RS comb used to spread the RS across the bandwidth such that every kcfrequency resource is occupied by the RS, mRSis the total number of RS across the frequency resource, and UE ports is the number of UE ports.

[0068] The inverse beam space transformation component 145 generates a denoised antennafrequency representation of the channel 150. The denoised antenna-frequency representation of the channel 150 may be used for various purposes, including but not limited to, channel equalization, beamforming configuration, channel state information (CSI) feedback, scheduling of frequency and time resources, designing precoders and / or combiners, and / or link adaptation.

[0069] Figure 2 is a diagram showing further details of the noise filtering component, according to some embodiments.

[0070] As shown in the diagram, the noise filtering component 130 may include a ML model 220. The noise filtering component 130 may obtain a noisy beam-delay representation of the channel 200 (e.g., from the delay domain transformation component 125, as described above). The noise filtering component 130 may also obtain a PE 210. The PE 210 may encode position information of beams and delay taps in the beam-delay domain. The noisy beam-delay representation of the channel 200 and the PE 210 may both have a dimension of beams x delay taps. The noise filtering component 130 may superimpose (represented in the diagram using the plus symbol) the PE 210 on the noisy beam-delay representation of the channel 200 in the beamdelay domain to generate an augmented beam-delay representation of the channel 215 (augmented with position information provided by the PE 210). The PE may be superimposed directly on the noisy beam-delay representation of the channel. This is different from conventional PE techniques where the input to the ML model is first embedded to a different dimension (e.g., tokenized) before being superimposed with the PE. The noise filtering component 130 may provide the augmented beam-delay representation of the channel 215 to the ML model 220. The ML model 220 may have been previously trained to denoise a noisy beamdelay representation of a channel using any suitable training data and any suitable training mechanism (e.g., trained with noisy representations of the channel (with the superimposition of the PE described herein below) and a supervised learning mechanism). The ML model 220 may output a denoised beam-delay representation of the channel 230. The denoised beam-delay representation of the channel 230 may be transformed to the antenna-frequency domain, as described above, so that it can be used for a desired purpose (e.g., channel equalization).

[0071] The PE may be designed to assist the ML model 220 with denoising the noisy beamdelay representation of the channel by encoding information regarding the relative positions of the dominant beam and delay taps in the channel. Furthermore, the PE may be designed such that most of the position information is encoded in a region the channel in the beam-delay domain where channel energy is expected to be sparse so that the superimposition of the PE does not interfere with the underlying channel pattern. In this way, the PE design takes thechannel properties into consideration. An example design of the PE is further described below to illustrate an embodiment.

[0072] The following description assumes that a noisy representation of the channel in the antenna-frequency domain has been transformed to the beam-delay domain such that most of the channel energy is concentrated on a few beams and delay taps and that delay compensation has been applied such that the channel energy is concentrated on the lower delay taps.

[0073] In an embodiment, the PE is designed such that the positional information of beams in the beam-delay domain is embedded along the delay tap dimension. Furthermore, the PE uniformly superimposes on the dominant delay taps (the delay taps that contain most of the channel energy), thereby helping to preserve the underlying channel pattern, while delegating most of the position information to the non-dominant delay taps.

[0074] For example, in an embodiment, the PE is generated according to equation 12:J k

[0075] (12) PE(b, t) = e s© = eja>tb

[0076] In equation 12, PE(Z>, t) is the PE and b E [0, ••• , B — 1] and t E [T — 1, ••• , 0] such that 1B and T are the number of beams and delay taps, respectively. Also, )t= —7- Zis the number 8 T) of delay taps, and <5 is a constant greater than zero (<5 > 0).

[0077] Equation 12 may be reformulated as follows:

[0078] (13) PE(b, t) = cos(mt6) + j sin(mtb)

[0079] It can be observed from equation 13, that the PE is given by cosine and sine waves superimposed on the real and imaginary beam-delay channel coefficients, respectively. Furthermore, the frequency of the waves, defined by mt, increases along the delay tap dimension for each beam. The rate of change of frequency depends on <5. This results in a geometric progression from 2nB to 2n on the wavelength, which effectively translates to the PE having the following properties along both the real and imaginary planes: (1) a slower change in the amplitude induced by the PE superimposed on the noisy beam-delay representation of the channel at the initial delay tap positions (lower delay tap indexes), where most of the channel energy is concentrated, thereby resulting in preservation of the underlying channel pattern; and (2) a faster change in the amplitude induced by the PE superimposed on the noisy beamdelay representation of the channel at the higher delay tap positions (higher delay tap indexes), thereby encoding most of the position information of beams in a region of the channel in the beam-delay domain where channel energy is expected to be sparse.

[0080] The above-mentioned properties help the ML model to learn the underlying channel pattern, while leveraging the position information encoded by the PE. It is noted that this PE design is different from the conventional PE design used in NLP, where frequency decreasesalong the embedding dimension (e.g., for Equation 11, t G [0, ••• , T — 1]). Applying the conventional PE design to the denoising context would result in the underlying pattern of beams and delay taps carrying the dominant channel energy being modified, making it difficult for the ML model to learn the underlying channel pattern.

[0081] In an alternative embodiment, the PE is defined in a reverse order with respect to the beam index, which results in flipping the encoding direction in the beam-delay domain. For example, the PE may be generated according to the below equation:

[0082] (14) PE(b, t) = cos(mt(B — b)) + j sin(mt(B — b))1

[0083] In equation 13, PE(b, f) is the PE, )t= — 6 is a constant greater than zero (<5 > 0), 8vr) and B is the number of beams.

[0084] Figures 3-5 show graph representations of an example PE, an example representation of a channel, and an example augmented representation of the channel that has been superimposed with the PE, respectively. In the examples, 6 is set to B (i.e., 6=B). Also, DMRS channel estimation is considered for a single layer such that the representation of the channel in the beam-delay domain has a dimension of [beams, taps, DMRS ports]=[ 128,312,1], It is noted that in this example the channel is assumed to be noise free to highlight the design principles of the PE design described herein. In the graphs shown in Figures 3-5, the x-axis represents the delay tap dimension and the y-axis represents the beam dimension.

[0085] Figure 3 is a diagram showing graph representations of the real and imaginary parts of a PE, according to some embodiments. As shown in the diagram, the real part of the PE 300A and the imaginary part of the PE 300B have waveforms (e.g., sinusoidal waves) that increase in frequency along the delay tap dimension (e.g., as depicted by fluctuations in shading in the diagram).

[0086] Figure 4 is a diagram showing graph representations of the real and imaginary parts of a representation of a channel in the beam-delay domain, according to some embodiments. As shown in the diagram, most of the channel energy in the real part of the representation of the channel 400A and the imaginary part of the representation of the channel 400A is concentrated on a few beams and the lower delay taps (e.g., as depicted by the shadings on the left side of the graphs). The shading represents the channel energy.

[0087] Figure 5 is a diagram showing graph representations of the real and imaginary parts of an augmented representation of the channel, according to some embodiments. As shown in the diagram, the real part of the augmented representation of the channel 500A is a superimposition of the real part of the PE on the real part of the beam-delay representation of the channel. Similarly, the imaginary part of the augmented representation of the channel 500B is asuperimposition of the imaginary part of the PE on the imaginary part of the beam-delay representation of the channel. As can be seen in the diagram, most of the position information is encoded in the higher delay taps (e.g., as depicted by the more rapid fluctuations in shading in the diagram), with a more uniform imposition of position information in the lower delay taps (e.g., as depicted by the less rapid fluctuations in shading in the diagram) that contain most of the channel energy.

[0088] Figure 6 is a diagram showing graph representations of the real and imaginary parts of a representation of the channel that is superimposed with a conventional PE, according to some embodiments. As can be seen in the diagram, most of the position information in the real part of the representation of the channel 600A and the imaginary part of the representation of the channel 600B is encoded in the lower delay taps that contain most of the channel energy. This results in the channel energy being modified, making it difficult for a ML model to learn the underlying channel pattern.

[0089] In the examples described above, the noisy antenna-frequency representation of the channel is transformed to the beam-delay domain in a manner where the channel energy is concentrated on the lower delay taps. Thus, the PE is designed to have a waveform that increases in frequency along the delay tap dimension as delay tap index increases. However, other embodiments may transform the noisy antenna-frequency representation of the channel to the beam-delay domain differently. For example, other embodiments may transform the noisy antenna-frequency representation of the channel to the beam-delay domain in a manner where the channel energy is concentrated on the higher delay taps. In such embodiments, the PE may be designed to have a waveform that increases in frequency along the delay tap dimension as delay tap index decreases. More generally, the noisy antenna-frequency representation of the channel may be transformed to the beam-delay domain in a manner where the channel energy is concentrated on delay taps located in a finite portion of the delay tap dimension. The PE may be designed to encode position information of beams and delay taps in the beam-delay domain using a waveform that increases in frequency along the delay tap dimension when going in a direction away from the finite portion of the delay tap dimension.

[0090] Figure 7 is a flow diagram of a method for denoising a representation of a wireless channel, according to some embodiments. The wireless channel may be a channel between a first wireless communication device and a second wireless communication device. In an embodiment, the method is performed by a computing device (e.g., a network device). In an embodiment, the computing device is the first wireless communication device or the second wireless communication device, but in other embodiments, the computing device is separate from the first wireless communication device and the second wireless communication device. Inan embodiment, the first wireless communication device is a base station in a mobile network and the second wireless communication device is a user equipment (UE) in the mobile network or vice versa. In an embodiment, different operations are performed by different components / entities (e.g., some of the operations are performed by a radio unit and other operations are performed by a baseband unit).

[0091] The operations in the flow diagram will be described with reference to the exemplary embodiments of the other figures. However, it should be understood that the operations of the flow diagram can be performed by embodiments other than those discussed with reference to the other figures, and the embodiments discussed with reference to these other figures can perform operations different than those discussed with reference to the flow diagram.

[0092] Also, while the flow diagrams in the figures show a particular order of operations performed by certain embodiments, it should be understood that such order is provided by way of example and not intended to be limiting (e.g., alternative embodiments may perform the operations in a different order, combine certain operations, overlap certain operations, etc.).

[0093] At operation 710, the computing device obtains a beam-delay representation of the wireless channel in a beam-delay domain in which channel energy is expected to be concentrated on delay taps located in a finite portion of a delay tap dimension. In an embodiment, the computing device obtains an antenna-frequency representation of the wireless channel in an antenna-frequency domain and transforms the antenna-frequency representation of the wireless channel from the antenna-frequency domain to the beam-delay domain to generate the beam-delay representation of the wireless channel. In an embodiment, the computing device analyzes a reference signal sent by the second wireless communication device to the first wireless communication device over the wireless channel to generate the antenna-frequency representation of the wireless channel. In an embodiment, the reference signal is any one of a CSI-RS, a SRS, and a DMRS.

[0094] At operation 720, the computing device obtains a positional encoding that encodes position information of beams and delay taps in the beam-delay domain using a waveform that increases in frequency along the delay tap dimension when going in a direction away from the finite portion of the delay tap dimension. In an embodiment, the finite portion of the delay tap dimension corresponds to a lowermost portion of the delay tap dimension, wherein the waveform increases in frequency along the delay tap dimension as delay tap index increases. In another embodiment, the finite portion of the delay tap dimension corresponds to a highermost portion of the delay tap dimension, wherein the waveform increases in frequency along the delay tap dimension as delay tap index decreases. In an embodiment, the positional encoding isgenerated according to equation 13. In an embodiment, the positional encoding is generated according to equation 14.

[0095] At operation 730, the computing device superimposes the positional encoding on the beam-delay representation of the wireless channel to generate an augmented beam-delay representation of the wireless channel.

[0096] At operation 740, the computing device applies a machine learning model to the augmented beam-delay representation of the wireless channel to generate a denoised beam-delay representation of the wireless channel. In an embodiment, the machine learning model is a neural network (e.g., a transformer). The machine learning model may have been trained to denoise representations of wireless channels using a supervised learning technique or other suitable learning technique.

[0097] In an embodiment, the computing device transforms the denoised beam-delay representation of the wireless channel from the beam-delay domain to an antenna-frequency domain to generate a denoised antenna-frequency representation of the wireless channel. In an embodiment, the denoised antenna-frequency representation of the wireless channel is used by a wireless communication device for performing one or more of: channel equalization, beamforming configuration, channel state information (CSI) feedback, scheduling of frequency and time resources, designing precoders and / or combiners, and link adaptation.

[0098] The denoising technique disclosed herein may provide better denoising compared to other denoising techniques. For example, it has been found that the denoising technique disclosed herein performs better at denoising a representation of a channel compared to each of (1) a legacy non-ML denoising technique (e.g., a denoising technique that is based on analyzing statistical properties of the channel); (2) a ML denoising technique that does not use PE (i.e., no PE is superimposed on the input to the ML model); and (3) a ML denoising technique that uses a PE design in which most of the position information is encoded in the region of the channel that contains most of the channel energy.

[0099] Various embodiments have been described herein that make use of ML and PE techniques to better denoise a representation of a channel. However, it should be appreciated that the ML and PE techniques disclosed herein can be applied to other applications and use cases. For example, the ML and PE techniques can be used for channel state information (CSI) compression and artificial intelligence (Al) receiver. As an example, for the CSI compression use case, the input data, which may be a representation of a channel in the antenna-frequency domain, may be transformed to the beam-delay domain to concentrate the channel power to a finite range. The PE design disclosed herein can then be superimposed on the beam-delay domain representation of the channel before the beam-delay domain representation of thechannel is provided to the ML model to help the ML model with training and inferencing for purposes of CSI compression. More generally, an embodiment is a method performed by a computing device to assist a machine learning model that takes a representation of a wireless channel between a first wireless communication device and a second wireless communication device as input. The method may include obtaining a beam-delay representation of the wireless channel in a beam-delay domain in which channel energy is expected to be concentrated on delay taps located in a finite portion of a delay tap dimension, generating obtaining a positional encoding that encodes position information of beams and delay taps in the beam-delay domain using a waveform that increases in frequency along the delay tap dimension when going in a direction away from the finite portion of the delay tap dimension, superimposing the positional encoding on the beam-delay representation of the wireless channel to generate an augmented beam-delay representation of the wireless channel, and applying the machine learning model to the augmented beam-delay representation of the wireless channel. In an embodiment, the machine learning model is trained to generate a denoised representation of the wireless channel. In an embodiment, the machine learning model is trained to perform channel state information (CSI) compression.

[0100] Figure 8 is a diagram showing an environment in which a representation of a channel can be denoised, according to some embodiments. As shown in the diagram, the environment includes network device (ND) 800 and ND 810.

[0101] Network device (ND) 800 may, in some embodiments, be an electronic device that can be communicatively connected to other electronic devices on the network (e.g., other network devices, user equipment devices (UEs), radio base stations, etc.). In certain embodiments, network device 800 may include radio access features that provide wireless radio network access to other electronic devices (for example a “radio access network device" may refer to such a network device) such as user equipment devices (UEs). For example, network device 800 may be a base station, such as eNodeB in Long Term Evolution (LTE), NodeB in Wideband Code Division Multiple Access (WCDMA) or other types of base stations, as well as a Radio Network Controller (RNC), a Base Station Controller (BSC), or other types of control nodes. As depicted in Figure 8, the example network device 800 comprises processor 801, memory 802, interface 803, and antenna 804. These components may work together to provide various network device functionality as disclosed herein.

[0102] Processor 801 may be a microprocessor, controller, microcontroller, central processing unit, digital signal processor, application specific integrated circuit, field programmable gate array, any other type of electronic circuitry, or any combination of one or more of the preceding. The processor 801 may comprise one or more processor cores. Inparticular embodiments, some or all of the channel estimation and / or denoising functionality described herein may be implemented by processor 801 executing software instructions, either alone or in conjunction with other network device 800 components, such as memory 802.

[0103] Memory 802 may store code (which is composed of software instructions and which is sometimes referred to as computer program code or a computer program) and / or data using non-transitory machine-readable (e.g., computer-readable) media, such as machine-readable storage media (e.g., magnetic disks, optical disks, solid state drives, read only memory (ROM), flash memory devices, phase change memory) and machine-readable transmission media (e.g., electrical, optical, radio, acoustical or other form of propagated signals - such as carrier waves, infrared signals). For instance, memory 802 may comprise non-volatile memory containing code to be executed by processor 801. Where memory802 is non-volatile, the code and / or data stored therein can persist even when the network device is turned off (when power is removed). In some instances, while network device 800 is turned on that part of the code that is to be executed by the processor(s) 801 may be copied from non-volatile memory into volatile memory (e.g., dynamic random access memory (DRAM), static random access memory (SRAM)) of network device 800. In an embodiment, memory 802 includes a channel estimation module 805 that includes computer code that when executed by the processor 801 causes the ND 800 to perform the channel estimation and / or denoising operations disclosed herein.

[0104] Interface 803 may be used in the wired and / or wireless communication of signaling and / or data to or from network device 800. For example, interface 803 may perform any formatting, coding, or translating to allow network device 800 to send and receive data whether over a wired and / or a wireless connection. In some embodiments, interface 803 may comprise radio circuitry capable of receiving data from other devices in the network over a wireless connection and / or sending data out to other devices via a wireless connection. This radio circuitry may include transmitted s), receiver(s), and / or transceiver(s) suitable for radiofrequency communication. The radio circuitry may convert digital data into a radio signal having the appropriate parameters (e.g., frequency, timing, channel, bandwidth, etc.). The radio signal may then be transmitted via antennas 804 to the appropriate recipient(s). In some embodiments, interface 803 may comprise network interface controlled s) (NICs), also known as a network interface card, network adapter, local area network (LAN) adapter or physical network interface. The NIC(s) may facilitate in connecting the network device 800 to other devices allowing them to communicate via wire through plugging in a cable to a physical port connected to a NIC. As explained above, in particular embodiments, processor 801 may represent part of interface 803, and some or all of the functionality described as being provided by interface XI 03 may be provided more specifically by processor 801.

[0105] The components of network device 800 are each depicted as separate boxes located within a single larger box for reasons of simplicity in describing certain aspects and features of network device 800 disclosed herein. In practice however, one or more of the components illustrated in the example network device 800 may comprise multiple different physical elements (e.g., interface 803 may comprise terminals for coupling wires for a wired connection and a radio transceiver for a wireless connection).

[0106] The solution described herein may be implemented in the network device 800 by means of a computer program comprising instructions which, when executed on at least one processor, cause the at least one processor to carry out the actions according to any of the above features and embodiments, where appropriate.

[0107] While the modules are illustrated as being implemented in software stored in memory 802, other embodiments implement part or all of each of these modules in hardware.

[0108] ND 810 may include similar components as ND 800 such as processors 811, memory 812, interface 813, and antennas 814. In an embodiment, similar to ND 800, memory 812 includes a channel estimation module 815 that includes computer code that when executed by the processor 811 causes the ND 810 to perform the channel estimation and / or denoising operations disclosed herein.

[0109] ND 800 and ND 810 may be wireless communication devices that can communicate with each other over a wireless channel. For example, in the example shown in the diagram, ND 800 is depicted as being a base station (e.g., a gNB) of a mobile network and ND 810 is depicted as being a mobile device (e.g., a UE) of the mobile network.

[0110] For a more thorough description of the example embodiment of network device 800 and 810 described in Figure 8, turn to Figure 9 and the associated description below.

[0111] Figure 9 illustrates two specific examples of how ND 800 may be implemented in certain embodiments of the described solution including: 1) a special-purpose network device 902 that uses custom processing circuits such as application-specific integrated-circuits (ASICs) and a proprietary operating system (OS); and 2) a general purpose network device 904 that uses common off-the-shelf (COTS) processors and a standard OS which has been configured to provide one or more of the features or functions disclosed herein.

[0112] Special-purpose network device 902 includes hardware 910 comprising processor(s) 912, and interface 916, as well as memory 918 having stored therein software 920. In one embodiment, the software 920 implements the modules described with regard to Figure 8. During operation, the software 920 may be executed by the hardware 910 to instantiate a set of one or more software instance(s) 922. Each of the software instance(s) 922, and that part of the hardware 910 that executes that software instance (be it hardware dedicated to that softwareinstance, hardware in which a portion of available physical resources (e.g., a processor core) is used, and / or time slices of hardware temporally shared by that software instance with others of the software instance(s) 922), form a separate virtual network element 930A-R. Thus, in the case where there are multiple virtual network elements 930A-R, each operates as one of the network devices from the preceding figures.

[0113] In an embodiment, software 920 includes code that when executed by the processing circuit 912 causes the special-purpose network device 902 to carry out the channel estimation and / or denoising operations disclosed herein.

[0114] Returning to Figure 9, the example general purpose network device 904 includes hardware 940 comprising a set of one or more processor(s) 942 (which are often COTS processors) and interface 946, as well as memory 948 having stored therein software 950. During operation, the processor(s) 942 execute the software 950 to instantiate one or more sets of one or more applications 964A-R. While certain embodiments do not implement virtualization, alternative embodiments may use different forms of virtualization. For example, in certain alternative embodiments virtualization layer 954 represents the kernel of an operating system (or a shim executing on a base operating system) that allows for the creation of multiple instances 962A-R called software containers that may each be used to execute one (or more) of the sets of applications 964A-R. In this embodiment, software containers 962A-R (also called virtualization engines, virtual private servers, or jails) are user spaces (typically a virtual memory space) that may be separate from each other and separate from the kernel space in which the operating system is run. In certain embodiments, the set of applications running in a given user space, unless explicitly allowed, may be prevented from accessing the memory of the other processes. In other such alternative embodiments virtualization layer 954 may represent a hypervisor (sometimes referred to as a virtual machine monitor (VMM)) or a hypervisor executing on top of a host operating system; and each of the sets of applications 964A-R may run on top of a guest operating system within an instance 962A-R called a virtual machine (which in some cases may be considered a tightly isolated form of software container that is run by the hypervisor). In certain embodiments, one, some or all of the applications are implemented as unikemel(s), which can be generated by compiling directly with an application only a limited set of libraries (e.g., from a library operating system (LibOS) including drivers / libraries of OS sendees) that provide the particular OS services needed by the application. As a unikemel can be implemented to run directly on hardware 940, directly on a hypervisor (in which case the unikemel is sometimes described as running within a LibOS virtual machine), or in a software container, embodiments can be implemented fully with unikemels running directly on a hypervisor represented by virtualization layer 954, unikemels running within softwarecontainers represented by instances 962A-R, or as a combination of unikemels and the abovedescribed techniques (e.g., unikemels and virtual machines both run directly on a hypervisor, unikemels and sets of applications that are run in different software containers).

[0115] The instantiation of the one or more sets of one or more applications 964A-R, as well as virtualization if implemented are collectively referred to as software instance(s) 952. Each set of applications 964 A-R, corresponding virtualization construct (e.g., instance 962 A-R) if implemented, and that part of the hardware 940 that executes them (be it hardware dedicated to that execution and / or time slices of hardware temporally shared by software containers 962A-R), forms a separate virtual network element(s) 960A-R.

[0116] The virtual network element(s) 960A-R perform similar functionality to the virtual network element(s) 930A-R. This virtualization of the hardware 940 is sometimes referred to as network function virtualization (NFV)). Thus, NFV may be used to consolidate many network equipment types onto industry standard high volume server hardware, physical switches, and physical storage, which could be located in for example data centers and customer premise equipment (CPE). However, different embodiments of the invention may implement one or more of the software container(s) 962A-R differently. While embodiments of the invention are illustrated with each instance 962A-R corresponding to one VNE 960A-R, alternative embodiments may implement this correspondence at a finer level granularity; it should be understood that the techniques described herein with reference to a correspondence of instances 962A-R to VNEs also apply to embodiments where such a finer level of granularity and / or unikemels are used.

[0117] In an embodiment, software 950 includes code that when executed by the processing circuit 942 causes the general purpose network device 904 to carry out the channel estimation and / or denoising operations disclosed herein.

[0118] The third exemplary ND implementation in Figure 9 is a hybrid network device 906, which includes both custom ASICs / proprietary OS and COTS processors / standard OS in a single ND or a single card within an ND. In certain embodiments of such a hybrid network device, a platform virtual machine (VM), such as a VM that that implements the functionality of the special-purpose network device 902, could provide for para-virtualization to the hardware present in the hybrid network device 906.

[0119] Some portions of the preceding detailed descriptions have been presented in terms of algorithms and symbolic representations of transactions on data bits within a computer memory. These algorithmic descriptions and representations are the ways used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. An algorithm is here, and generally, conceived to be a self-consi stent sequence of transactionsleading to a desired result. The transactions are those requiring physical manipulations of physical quantities. Usually, though not necessarily, these quantities take the form of electrical or magnetic signals capable of being stored, transferred, combined, compared, and otherwise manipulated. It has proven convenient at times, principally for reasons of common usage, to refer to these signals as bits, values, elements, symbols, characters, terms, numbers, or the like.

[0120] It should be borne in mind, however, that all of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless specifically stated otherwise as apparent from the above discussion, it is appreciated that throughout the description, discussions utilizing terms such as "processing" or "computing" or "calculating" or "determining" or "displaying" or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system memories or registers or other such information storage, transmission or display devices.

[0121] The algorithms and displays presented herein are not inherently related to any particular computer or other apparatus. Various general purpose systems may be used with programs in accordance with the teachings herein, or it may prove convenient to construct more specialized apparatus to perform the required method transactions. The required structure for a variety of these systems will appear from the description above. In addition, embodiments are not described with reference to any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the teachings of embodiments as described herein.

[0122] An embodiment may be an article of manufacture in which a non-transitory machine- readable storage medium (such as microelectronic memory) has stored thereon instructions (e.g., computer code) which program one or more data processing components (generically referred to here as a “processor”) to perform the operations described above. In other embodiments, some of these operations might be performed by specific hardware components that contain hardwired logic (e.g., dedicated digital filter blocks and state machines). Those operations might alternatively be performed by any combination of programmed data processing components and fixed hardwired circuit components.

[0123] Throughout the description, embodiments have been presented through flow diagrams. It will be appreciated that the order of transactions and transactions described in these flow diagrams are only intended for illustrative purposes and not intended to be limiting. Onehaving ordinary skill in the art would recognize that variations can be made to the flow diagrams.

[0124] In the foregoing specification, embodiments have been described with reference to specific example embodiments thereof. It will be evident that various modifications may be made thereto without departing from the broader spirit and scope of the disclosure provided herein. The specification and drawings are, accordingly, to be regarded in an illustrative sense rather than a restrictive sense.

Claims

CLAIMSWhat is claimed is:

1. A method performed by a computing device to denoise a representation of a wireless channel between a first wireless communication device and a second wireless communication device, the method comprising: obtaining (710) a beam-delay representation of the wireless channel in a beam-delay domain in which channel energy is expected to be concentrated on delay taps located in a finite portion of a delay tap dimension; obtaining (720) a positional encoding that encodes position information of beams and delay taps in the beam-delay domain using a waveform that increases in frequency along the delay tap dimension when going in a direction away from the finite portion of the delay tap dimension; superimposing (730) the positional encoding on the beam-delay representation of the wireless channel to generate an augmented beam-delay representation of the wireless channel; and applying (740) a machine learning model to the augmented beam-delay representation of the wireless channel to generate a denoised beam-delay representation of the wireless channel.

2. The method of claim 1, further comprising: obtaining an antenna-frequency representation of the wireless channel in an antennafrequency domain; and transforming the antenna-frequency representation of the wireless channel from the antenna-frequency domain to the beam-delay domain to generate the beam-delay representation of the wireless channel.

3. The method of claim 2, further comprising: analyzing a reference signal sent by the second wireless communication device to the first wireless communication device over the wireless channel to generate the antenna-frequency representation of the wireless channel.

4. The method of claim 3, wherein the reference signal is any one of: a channel state information reference signal (CSI-RS), a sounding reference signal (SRS), and a demodulation reference signal (DMRS).

5. The method of claim 1, wherein the finite portion of the delay tap dimension corresponds to a lowermost portion of the delay tap dimension, wherein the waveform increases in frequency along the delay tap dimension as delay tap index increases.

6. The method of claim 1, wherein the finite portion of the delay tap dimension corresponds to a highermost portion of the delay tap dimension, wherein the waveform increases in frequency along the delay tap dimension as delay tap index decreases.

7. The method of claim 1, wherein the positional encoding is generated according to the following equation:PE(b, t) = cos(mtb) + j sin(mtb) wherein PE(b, / ) is the positional encoding, b£[0, • • -,B-1 ], tG[Z-l,---,0], B is a number of beams, T is a number of delay taps, )t= — - and 6 is a constant greater than zero.

8. The method of claim 1, wherein the positional encoding is generated according to the following equation:PE(b, t) = cos(mt(B — b)) + j sin(mt(B — b)) wherein PE(b, / ) is the positional encoding, bG[0,---, -l], tG[Z-l,---,0], B is a number of beams, T is a number of delay taps, )t= — - and 6 is a constant greater than zero.

9. The method of claim 1, further comprising: transforming the denoised beam-delay representation of the wireless channel from the beam-delay domain to an antenna-frequency domain to generate a denoised antenna-frequency representation of the wireless channel.

10. The method of claim 9, wherein the denoised antenna-frequency representation of the wireless channel is used by the first wireless communication device for performing one or more of: channel equalization, beamforming configuration, channel state information (CSI) feedback, scheduling of frequency and time resources, designing precoders and / or combiners, and link adaptation.

11. The method of claim 1, wherein the computing device is the first wireless communication device or the second wireless communication device.

12. The method of claim 1, wherein the first wireless communication device is a base station in a mobile network and the second wireless communication device is a user equipment (UE) in the mobile network.

13. The method of claim 1, wherein the machine learning model is a neural network.

14. A computing device (904) to denoise a representation of a wireless channel between a first wireless communication device and a second wireless communication device, the computing device comprising: one or more processors (942); and a non-transitory machine-readable storage medium (948) that stores instructions, which when executed by the one or more processors, causes the computing device to perform the method steps of any one of claims 1-13.

15. A machine-readable medium comprising computer program code which when executed by a computing device carries out the method steps of any of claims 1-13.

16. A method performed by a computing device to assist a machine learning model that takes a representation of a wireless channel between a first wireless communication device and a second wireless communication device as input, the method comprising: obtaining a beam-delay representation of the wireless channel in a beam-delay domain in which channel energy is expected to be concentrated on delay taps located in a finite portion of a delay tap dimension; generating obtaining a positional encoding that encodes position information of beams and delay taps in the beam-delay domain using a waveform that increases in frequency along the delay tap dimension when going in a direction away from the finite portion of the delay tap dimension; superimposing the positional encoding on the beam-delay representation of the wireless channel to generate an augmented beam-delay representation of the wireless channel; and applying the machine learning model to the augmented beam-delay representation of the wireless channel.

17. The method of claim 16, wherein the machine learning model is trained to generate a denoised representation of the wireless channel.

18. The method of claim 16, wherein the machine learning model is trained to perform channel state information (CSI) compression.

19. A computing device (904) to assist a machine learning model that takes a wireless channel between a first wireless communication device and a second wireless communication device as input, the computing device comprising: one or more processors (942); and a non-transitory machine-readable storage medium (948) that stores instructions, which when executed by the one or more processors, causes the computing device to perform the method steps of any one of claims 16-18.

20. A machine-readable medium comprising computer program code which when executed by a computing device carries out the method steps of any of claims 16-18.

Images