Encoding techniques for neural network architectures

Abstract

Methods, systems, and devices for wireless communication are described. A user equipment (UE) can receive an indication of one or more encoding operations for encoding a compressed data set, the one or more encoding operations including a differential encoding operation or an entropy encoding operation or both. In some examples, using a neural network, the UE can first encode the data set based on an additional encoding operation to generate a compressed data set, and then quantize the compressed data set encoded based on the additional encoding operation. Subsequently, after the data set has been initially encoded and then quantized, the UE can further encode and compress the data set using the indication of the one or more encoding operations. The UE can then transmit the data set to a second device based on the one or more encoding operations.

Description

[0001] Cross-references to related applications

[0002] This patent application claims the benefit of U.S. Patent Application No. 17 / 194,077, filed March 5, 2021, entitled “ENCODING TECHNIQUES FOR NEURAL NETWORK ARCHITECTURES”, which has been assigned to its assignee. Technical Field

[0003] The following pertains to wireless communication, including coding techniques used in neural network architectures. Background Technology

[0004] Wireless communication systems are widely deployed to provide various types of communication content, such as voice, video, packet data, messaging, broadcasting, and so on. These systems are able to support communication with multiple users by sharing 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-A 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 Spectrum Orthogonal Frequency Division Multiplexing (DFT-S-OFDM).

[0005] A wireless multiple access communication system may include one or more base stations or one or more network access nodes, each of which simultaneously supports communication from multiple communication devices, also referred to as user equipment (UEs). In some wireless communication systems, the UE may perform several channel condition measurements as part of its communication with the base station. In some examples, these measurements may generate a considerable amount of data to be sent to the base station to assist in network management. Summary of the Invention

[0006] The described techniques relate to methods, systems, devices, and apparatuses that support improvements in encoding techniques for neural network architectures. Typically, the described techniques provide a user equipment (UE) with instructions for one or more encoding operations to encode a compressed dataset, including differential encoding operations, entropy encoding operations, or both. In some examples, using a neural network, the UE may first encode the dataset based on additional encoding operations (e.g., a single-shot encoder) to generate a compressed dataset, and then quantize the compressed dataset encoded based on the additional encoding operations. Subsequently, after the dataset has been initially encoded and then quantized, the UE may use instructions for one or more encoding operations to further encode and compress the dataset. The UE may then transmit the dataset (e.g., an encoded, quantized, and compressed dataset) to a second device (e.g., a base station, an additional UE, etc.) based on the use of one or more encoding operations to further encode information after using the neural network and quantizing the data. In some examples, differential coding operations may include encoding a certain amount of data in a compressed dataset based on previous values ​​of a certain amount of data (such as initial values, initial reconstructed values, previous reconstructed values, previous values ​​of the same data from previous time instances, etc.) (e.g., after an additional coding operation).

[0007] A method for wireless communication at a UE is described. The method may include: receiving an instruction for encoding one or more encoding operations on a compressed dataset, the one or more encoding operations including differential coding or entropy coding or both; encoding the dataset by a neural network to generate a compressed dataset; quantizing the compressed dataset encoded by the neural network; encoding the quantized and compressed dataset based on the received instruction for the one or more encoding operations; and transmitting the encoded, quantized, and compressed dataset to a second device after encoding the compressed dataset based on the one or more encoding operations.

[0008] An apparatus for wireless communication at a UE is described. The apparatus may include a processor, a memory coupled to the processor, and instructions stored in the memory. The instructions are executable by the processor to cause the apparatus to: receive instructions for encoding one or more encoding operations on a compressed dataset, the one or more encoding operations including differential coding or entropy coding or both; encode the dataset by a neural network to generate a compressed dataset; quantize the compressed dataset encoded by the neural network; encode the quantized and compressed dataset based on the received instructions for the one or more encoding operations; and transmit the encoded, quantized, and compressed dataset to a second device after encoding the compressed dataset based on the one or more encoding operations.

[0009] Another apparatus for wireless communication at a UE is described. The apparatus may include: components for receiving instructions for one or more encoding operations for encoding a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both; components for encoding the dataset by a neural network to generate a compressed dataset; components for quantizing the compressed dataset encoded by the neural network; components for encoding the quantized and compressed dataset based on the received instructions for one or more encoding operations; and components for transmitting the encoded, quantized, and compressed dataset to a second device after encoding the compressed dataset based on one or more encoding operations.

[0010] A non-transitory computer-readable medium is described, storing code for wireless communication at a UE. The code may include instructions executable by a processor to perform the following operations: receiving instructions for one or more encoding operations, including differential coding or entropy coding or both, of a compressed dataset; encoding the dataset by a neural network to generate a compressed dataset; quantizing the compressed dataset encoded by the neural network; encoding the quantized and compressed dataset based on the received instructions for one or more encoding operations; and transmitting the encoded, quantized, and compressed dataset to a second device after encoding the compressed dataset based on the one or more encoding operations.

[0011] In some examples of the methods, apparatuses and nontransitory computer-readable media described herein, an instruction to receive one or more encoded operations may include an operation, feature, component or instruction for receiving one or more parameters corresponding to one or more encoded operations, each of the one or more parameters corresponding to a corresponding encoded operation of the one or more encoded operations, wherein the quantized and compressed dataset may be encoded based on one or more parameters.

[0012] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, encoding a quantized and compressed dataset may include operations, features, components, or instructions for encoding the quantized and compressed dataset using differential coding operations after encoding the dataset using a neural network.

[0013] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, encoding a quantized and compressed dataset may include operations, features, components, or instructions for encoding the quantized and compressed dataset using an entropy encoding operation after the dataset has been encoded using a neural network.

[0014] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, encoding a quantized and compressed dataset may include operations, features, components, or instructions for determining a difference between a first value of data at an initial time instance and a second value of data at a second time instance after the initial time instance, following the quantization of the compressed dataset encoded by a neural network. The difference may be determined based on instructions for one or more encoding operations, and the quantized and compressed dataset may be encoded based on the difference.

[0015] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, encoding a quantized and compressed dataset may include operations, features, components, or instructions for determining a difference between a first reconstructed value of data at a first time instance and a second reconstructed value of data at a second time instance following the first time instance, after quantizing the compressed dataset encoded by a neural network, wherein the difference value may be determined based on instructions of one or more encoding operations, and the quantized and compressed dataset may be encoded based on the difference value.

[0016] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, encoding a quantized and compressed dataset may include operations, features, components, or instructions for determining, after quantizing the compressed dataset encoded by a neural network, an initial reconstructed value of the data at an initial time instance associated with the encoded dataset, and for determining, after quantization, a difference between the additional reconstructed value of the data at an additional time instance following the initial time instance and the initial reconstructed value of the data, wherein the difference may be determined based on instructions of one or more encoding operations, and the quantized and compressed dataset may be encoded based on the difference.

[0017] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, differential coding operations may include encoding a quantity of data based on previous values ​​of a quantity of data in a compressed dataset.

[0018] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, entropy coding operations may include encoding a compressed dataset using one or more symbols of varying lengths based on the probability of symbol occurrence.

[0019] A method for wireless communication at a device is described. The method may include: sending to a UE an instruction for one or more encoding operations by the UE to encode a compressed dataset, the one or more encoding operations including differential coding or entropy coding or both; receiving from the UE an encoded, quantized, and compressed dataset that has been encoded based on the one or more encoding operations after a quantization operation; decoding the encoded, quantized, and compressed dataset based on the one or more encoding operations to generate a compressed dataset; and generating the dataset by decoding the compressed dataset based on the decoding of the encoded, quantized, and compressed dataset using a neural network.

[0020] An apparatus for wireless communication at a device is described. The apparatus may include a processor, a memory coupled to the processor, and instructions stored in the memory. The instructions are executable by the processor to cause the apparatus to: send to a UE an instruction for the UE to encode one or more encoding operations, the one or more encoding operations including differential coding operations or entropy coding operations or both; receive from the UE an encoded, quantized, and compressed dataset having been encoded based on the one or more encoding operations after a quantization operation; decode the encoded, quantized, and compressed dataset based on the one or more encoding operations to generate a compressed dataset; and generate a dataset by decoding the compressed dataset by a neural network based on the decoding of the encoded, quantized, and compressed dataset.

[0021] Another apparatus for wireless communication at a device is described. The apparatus may include: components for transmitting to a UE an instruction for one or more encoding operations performed by the UE to encode a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both; components for receiving from the UE an encoded, quantized, and compressed dataset after encoding the compressed dataset based on the one or more encoding operations following a quantization operation; components for decoding the encoded, quantized, and compressed dataset based on the one or more encoding operations to generate a compressed dataset; and components for decoding the compressed dataset by a neural network based on decoding the encoded, quantized, and compressed dataset to generate a dataset.

[0022] A non-transitory computer-readable medium is described, storing code for wireless communication at a device. The code may include instructions executable by a processor to perform the following operations: sending to a UE one or more encoding operations for encoding a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both; receiving from the UE an encoded, quantized, and compressed dataset after encoding the compressed dataset based on one or more encoding operations following a quantization operation; decoding the encoded, quantized, and compressed dataset based on one or more encoding operations to generate a compressed dataset; and generating a dataset by decoding the compressed dataset based on the decoding of the encoded, quantized, and compressed dataset by a neural network.

[0023] In some examples of the methods, apparatuses and nontransitory computer-readable media described herein, an instruction to send one or more encoded operations may include an operation, feature, component or instruction for sending one or more parameters corresponding to one or more encoded operations, each of the one or more parameters corresponding to a corresponding encoded operation of the one or more encoded operations, wherein the quantized and compressed dataset may be encoded based on one or more parameters.

[0024] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, decoding an encoded, quantized, and compressed dataset may include operations, features, components, or instructions for decoding the encoded, quantized, and compressed dataset using differential decoding operations prior to decoding the dataset using a neural network.

[0025] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, decoding an encoded, quantized, and compressed dataset may include operations, features, components, or instructions for decoding the encoded, quantized, and compressed dataset using an entropy decoding operation prior to decoding the dataset using a neural network.

[0026] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, receiving an encoded, quantized, and compressed dataset may include operations, features, components, or instructions for receiving the encoded, quantized, and compressed dataset, which includes differential values ​​of data in the dataset, which may be based on initial values ​​of the data.

[0027] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, receiving an encoded, quantized, and compressed dataset may include operations, features, components, or instructions for receiving the encoded, quantized, and compressed dataset, which includes difference values ​​of data in the dataset, which may be based on previously reconstructed values ​​of the data.

[0028] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, receiving an encoded, quantized, and compressed dataset may include operations, features, components, or instructions for receiving the encoded, quantized, and compressed dataset, which includes difference values ​​of data in the dataset, which may be based on initial reconstructed values ​​of the data.

[0029] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, differential decoding operations may include encoding a quantity of data based on previous values ​​of a quantity of data in a compressed dataset.

[0030] In some examples of the methods, apparatuses, and non-transitory computer-readable media described herein, entropy decoding operations may include encoding a compressed dataset using one or more symbols of varying lengths based on the probability of symbol occurrence. Attached Figure Description

[0031] Figure 1 An example of a wireless communication system that supports coding techniques for neural network architectures according to various aspects of this disclosure is illustrated.

[0032] Figure 2 An example of a wireless communication system that supports coding techniques for neural network architectures according to various aspects of this disclosure is illustrated.

[0033] Figure 3A and 3B An example of a compression program that supports encoding techniques for neural network architectures according to various aspects of this disclosure is illustrated.

[0034] Figure 4 An example of a compression configuration that supports encoding techniques for neural network architectures according to various aspects of this disclosure is illustrated.

[0035] Figure 5 Examples of compression and encoding configurations for encoding techniques used in neural network architectures, in accordance with various aspects of this disclosure, are illustrated.

[0036] Figure 6 Examples of compression and encoding configurations for encoding techniques used in neural network architectures, in accordance with various aspects of this disclosure, are illustrated.

[0037] Figure 7 An example of a processing flow for encoding techniques used in neural network architectures, according to various aspects of this disclosure, is illustrated.

[0038] Figure 8 and 9 A block diagram of a device supporting coding techniques for neural network architectures according to various aspects of this disclosure is shown.

[0039] Figure 10 A block diagram of a communication manager supporting coding techniques for neural network architectures according to various aspects of this disclosure is shown.

[0040] Figure 11 A diagram of a system including a device supporting coding techniques for neural network architectures is shown according to various aspects of this disclosure.

[0041] Figure 12 and 13 A block diagram of a device supporting coding techniques for neural network architectures according to various aspects of this disclosure is shown.

[0042] Figure 14 A block diagram of a communication manager supporting coding techniques for neural network architectures according to various aspects of this disclosure is shown.

[0043] Figure 15 A diagram of a system including a device supporting coding techniques for neural network architectures is shown according to various aspects of this disclosure.

[0044] Figures 16 to 20 A flowchart illustrating a method for encoding techniques used in neural network architectures according to various aspects of this disclosure is shown. Detailed Implementation

[0045] In some wireless communication systems, user equipment (UE) can perform several channel condition measurements as part of communication with the base station. For example, measurements may include per-antenna-port measurements of the channel and interference (e.g., channel state feedback), power measurements from the serving cell and neighboring cells, radio access technology (RAT) measurements (e.g., from Wi-Fi networks), sensor measurements, and so on. These measurements can generate a considerable amount of data to be transmitted to the base station to assist in network management. For example, a wireless communication system using an antenna panel with multiple elements can generate more measurement information than a wireless communication system using a single antenna or a smaller antenna panel with fewer elements. In some examples, the UE can use neural networks to compress measurements to reduce the size of the transmission carrying the measurements. In addition to compression techniques using neural networks, techniques for applying additional compression techniques to the measurement data are also needed.

[0046] A technique for compressing measurement data by a UE for transmission is described. The UE can measure one or more channel conditions. The UE can then use a neural network to compress the measurement data. The output of the neural network can be quantized to provide more transmittable data. In addition to neural network compression, the UE can use additional compression layers before sending the transmission to the base station. For example, the UE can use differential coding, entropy coding, or both to further compress the measurement data after compressing it with the neural network and before sending the transmission to the base station. Differential coding can be an example of encoding the difference between two measurements rather than the absolute value of the measurements. For example, differential coding can use a previous value, an initial value, a reconstructed value, an initial reconstructed value, or a combination thereof to indicate the differential value. In some examples, the UE can perform differential coding, entropy coding, or both to further compress the data after a quantization step associated with neural network processing. Additionally or alternatively, differential coding can occur at the input stage of the compression operation, where bits representing the difference between measurements are initially encoded and then input to the neural network for processing. The base station can then use differential decoding, entropy decoding, or both when decoding the compressed, encoded transmission from the UE. In some examples, the base station may send indications of differential coding and entropy coding (e.g., parameters for each coding) for use by the UE. Thus, coding operations performed by the transmitting device (e.g., the UE) and decoding operations performed by the receiving device (e.g., the base station) can be coordinated.

[0047] The aspects of this disclosure are initially described in the context of wireless communication systems. Additionally, aspects of this disclosure are illustrated by additional wireless communication systems, compression procedures, compression configurations, compression and encoding configurations, and processing flows. The aspects of this disclosure are further illustrated and described with reference to apparatus diagrams, system diagrams, and flowcharts related to encoding techniques used in neural network architectures.

[0048] Figure 1 An example of a wireless communication system 100 supporting coding techniques for neural network architectures according to various aspects of this disclosure is illustrated. The wireless communication system 100 may include one or more base stations 105, one or more UEs 115, and a core network 130. In some examples, the wireless communication system 100 may be a Long Term Evolution (LTE) network, an Advanced LTE (LTE-A) network, an LTE-A Pro network, or a New Radio (NR) network. In some examples, the wireless communication system 100 may support enhanced broadband communication, ultra-reliable (e.g., mission-critical) communication, low-latency communication, communication with low-cost and low-complexity devices, or any combination thereof.

[0049] Base stations 105 can be distributed throughout a geographical area to form a wireless communication system 100, and can be devices of different forms or with different capabilities. Base stations 105 and UE 115 can communicate wirelessly via one or more communication links 125. Each base station 105 can provide a coverage area 110, and UE 115 and base station 105 can establish one or more communication links 125 on the coverage area 110. The coverage area 110 can be an example of a geographical area where base station 105 and UE 115 can support signal communication according to one or more radio access technologies.

[0050] UE 115 can be distributed throughout the coverage area 110 of the wireless communication system 100, and each UE 115 can be stationary, mobile, or both at different times. UE 115 can be devices of different forms or with different capabilities. Figure 1 The diagram illustrates some example UE 115s. The UE 115 described herein is capable of communicating with various types of devices, such as other UE 115s, base station 105, or network devices (e.g., core network nodes, relay devices, integrated access and backhaul (IAB) nodes, or other network devices), such as... Figure 1 As shown in the image.

[0051] Base station 105 may communicate with core network 130, or communicate with each other, or both. For example, base station 105 may interface with core network 130 via one or more backhaul links 120 (e.g., via S1, N2, N3, or other interfaces). Base station 105 may communicate with each other directly (e.g., directly between base stations 105) or indirectly (e.g., via core network 130) or both via backhaul links 120 (e.g., via X2, Xn, or other interfaces). In some examples, backhaul link 120 may be or include one or more radio links.

[0052] One or more base stations 105 described herein may include, or may be referred to by those skilled in the art as, base station, radio base station, access point, radio transceiver, NodeB, eNodeB (eNB), next-generation NodeB or giga-NodeB (any of which may be referred to as gNB), home NodeB, home eNodeB or other suitable terms.

[0053] UE 115 may include or be referred to as a mobile device, wireless device, remote device, handheld device, or subscriber device, or some other suitable term, wherein "device" may also be referred to as a unit, station, terminal, or client, etc. UE 115 may also include or be referred to as a personal electronic device, such as a cellular phone, personal digital assistant (PDA), tablet computer, laptop computer, or personal computer. In some examples, UE 115 may include or be referred to as a wireless local loop (WLL) station, Internet of Things (IoT) device, Internet of Everything (IoE) device, or machine-type communication (MTC) device, etc., which may be implemented in various objects such as appliances or vehicles, meters, etc.

[0054] The UE 115 described in this document can communicate with various types of devices, such as other UE 115s that can sometimes act as relays, as well as base station 105 and network devices, including macro eNBs or gNBs, small cell eNBs or gNBs, or relay base stations, etc. Figure 1 As shown in the image.

[0055] UE 115 and base station 105 can communicate wirelessly with each other via one or more communication links 125 on one or more carriers. The term "carrier" can refer to a collection of radio frequency spectrum resources having a defined physical layer structure for supporting communication link 125. For example, a carrier for communication link 125 may include a portion of the radio spectrum band (e.g., a bandwidth portion (BWP)) operating according to one or more physical layer channels of a given radio access technology (e.g., LTE, LTE-A, LTE-A Pro, NR). Each physical layer channel may carry acquisition signaling (e.g., synchronization signals, system information), control signaling coordinating carrier operation, user data, or other signaling. Wireless communication system 100 can use carrier aggregation or multi-carrier operation to support communication with UE 115. Depending on the carrier aggregation configuration, UE 115 may be configured with multiple downlink component carriers and one or more uplink component carriers. Carrier aggregation can be used in conjunction with both frequency division duplex (FDD) and time division duplex (TDD) component carriers.

[0056] In some examples (e.g., in a carrier aggregation configuration), the carrier may also have acquisition signaling or control signaling that coordinates the operation of other carriers. The carrier may be associated with a frequency channel (e.g., an Evolved Universal Mobile Telecommunications System Terrestrial Radio Access (E-UTRA) Absolute Radio Frequency Channel Number (EARFCN)) and can be located according to a channel grid for discovery by UE115. The carrier can operate in standalone mode, where initial acquisition and connection can be performed by UE115 via the carrier, or the carrier can operate in non-standalone mode, where the connection is anchored using different carriers (e.g., carriers of the same or different radio access technologies).

[0057] The communication link 125 shown in the wireless communication system 100 may include uplink transmission from UE 115 to base station 105, or downlink transmission from base station 105 to UE 115. A carrier may carry either downlink or uplink communication (e.g., in FDD mode), or may be configured to carry both downlink and uplink communication (e.g., in TDD mode).

[0058] A carrier can be associated with a specific bandwidth of the radio spectrum, and in some examples, the carrier bandwidth can be referred to as the carrier or the “system bandwidth” of the wireless communication system 100. For example, the carrier bandwidth can be one of a number of bandwidths determined for a carrier of a specific radio access technology (e.g., 1.4, 3, 5, 10, 15, 20, 40, or 80 MHz). Devices of the wireless communication system 100 (e.g., base station 105, UE 115, or both) can have a hardware configuration that supports communication over a specific carrier bandwidth, or can be configured to support communication over one carrier bandwidth in a set of carrier bandwidths. In some examples, the wireless communication system 100 may include base station 105 or UE 115 that supports simultaneous communication via carriers associated with multiple carrier bandwidths. In some examples, each served UE 115 can be configured to operate on a portion (e.g., a subband, BWP) or all of the carrier bandwidth.

[0059] The signal waveform transmitted on a carrier can consist of multiple subcarriers (e.g., using multi-carrier modulation (MCM) techniques, such as orthogonal frequency division multiplexing (OFDM) or discrete Fourier transform extended OFDM (DFT-S-OFDM)). In a system employing MCM, a resource element can consist of one symbol period (e.g., the duration of a modulation symbol) and one subcarrier, where the symbol period and subcarrier spacing are inversely related. The 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). Therefore, the more resource elements the UE 115 receives and the higher the order of the modulation scheme, the higher the data rate the UE 115 can achieve. Wireless communication resources can refer to a combination of radio spectrum resources, time resources, and spatial resources (e.g., spatial layers or beams), and the use of multiple spatial layers can further improve the data rate or data integrity of communication with the UE 115.

[0060] One or more parameter sets for a carrier can be supported, where the parameter sets may include subcarrier spacing (Δf) and cyclic prefix. A carrier can be divided into one or more BWPs with the same or different parameter sets. In some examples, the UE115 can be configured with multiple BWPs. In some examples, a single BWP for a carrier can be active at a given time, and the UE115's communication can be limited to one or more active BWPs.

[0061] The time interval of base station 105 or UE 115 can be represented by a multiple of the basic time unit. For example, the basic time unit can refer to the sampling period T. s =1 / (Δf) max ·N f ) seconds, where Δf max This can represent the maximum supported subcarrier spacing, and N f This can represent the maximum supported Discrete Fourier Transform (DFT) size. Communication resources can be organized into time intervals based on 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).

[0062] 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 be divided (e.g., in the time domain) 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 pre-added to each symbol period). In some wireless communication systems 100, time slots may be further divided into multiple micro-time slots containing one or more symbols. In addition to the cyclic prefix, each symbol period may contain one or more (e.g., N) f Sampling period. The duration of the symbol period can depend on the subcarrier spacing or the operating frequency band.

[0063] 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)).

[0064] Physical channels can be multiplexed on a carrier using various techniques. For example, one or more of Time Division Multiplexing (TDM), Frequency Division Multiplexing (FDM), or hybrid TDM-FDM techniques can be used to multiplex physical control channels and physical data channels on a downlink carrier. The control region (e.g., a control resource set (CORESET)) of a physical control channel can be defined by a number of symbol periods and can be extended across the system bandwidth or a subset of the system bandwidth of the carrier. One or more control regions (e.g., CORESETs) can be configured for a set of UEs 115. For example, one or more UEs 115 can monitor or search for control information within a control region based on one or more search space sets, and each search space set can include one or more control channel candidates at one or more aggregation levels arranged in a cascaded manner. The aggregation level of control channel candidates can refer to the number of control channel resources (e.g., control channel elements (CCEs)) associated with encoded information having a control information format with a given payload size. The search space set can include a common search space set configured to send control information to multiple UEs 115, and a UE-specific search space set for sending control information to a specific UE 115.

[0065] Each base station 105 may provide communication coverage via one or more cells, such as macro cells, small cells, hotspots, or other types of cells, or any combination thereof. The term "cell" may refer to a logical communication entity used to communicate with base station 105 (e.g., via a carrier) and may be associated with an identifier used to distinguish neighboring cells (e.g., Physical Cell Identifier (PCID), Virtual Cell Identifier (VCID), etc.). In some examples, a cell may also refer to a geographic coverage area 110 or a portion of geographic coverage area 110 (e.g., a sector) on which a logical communication entity operates. Depending on various factors, such as the capabilities of base station 105, the range of these cells can vary from small areas (e.g., buildings, subsets of buildings) to large areas. For example, a cell may be or may include buildings, subsets of buildings, or external space between or overlapping with geographic coverage areas 110, etc.

[0066] Macro cells typically cover a relatively large geographic area (e.g., a radius of several kilometers) and can allow unrestricted access for UE 115 with a service subscription from a network provider supporting the macro cell. In contrast, small cells can be associated with a lower-power base station 105 and can operate in the same or different (e.g., licensed, unlicensed) frequency bands as macro cells. Small cells can provide unrestricted access to UE 115 with a service subscription from a network provider, or restricted access to UE 115 associated with that small cell (e.g., UE 115 in a Closed Subscriber Group (CSG), UE 115 associated with a user in a home or office). Base station 105 can support one or more cells and can also use one or more component carriers to support communication on one or more cells.

[0067] In some examples, a carrier can support multiple cells and can be configured with different cells based on different protocol types that can provide access for different types of devices (e.g., MTC, Narrowband IoT (NB-IoT), Enhanced Mobile Broadband (eMBB)).

[0068] In some examples, base station 105 may be mobile, thus providing communication coverage for mobile geographic coverage areas 110. In some examples, different geographic coverage areas 110 associated with different technologies may overlap, but the different geographic coverage areas 110 may be supported by the same base station 105. In other examples, overlapping geographic coverage areas 110 associated with different technologies may be supported by different base stations 105. Wireless communication system 100 may include, for example, a heterogeneous network, in which different types of base stations 105 use the same or different radio access technologies to provide coverage for various geographic coverage areas 110.

[0069] The wireless communication system 100 can support synchronous or asynchronous operation. For synchronous operation, base stations 105 can have similar frame timings, and transmissions from different base stations 105 can be approximately time-aligned. For asynchronous operation, base stations 105 can have different frame timings, and in some examples, transmissions from different base stations 105 can be time-disaligned. The techniques described herein can be used for both synchronous and asynchronous operation.

[0070] Some UE 115 devices, such as MTC or IoT devices, can be low-cost or low-complexity devices and can provide automated communication between machines (e.g., via machine-to-machine (M2M) communication). M2M communication or MTC can refer to data communication technologies that allow devices to communicate with each other or with base station 105 without human intervention. In some examples, M2M communication or MTC may include communication from devices that integrate sensors or meters to measure or capture information and relay such information to a central server or application, which uses the information or presents it to people interacting with the application. Some UE 115 devices can be designed to collect information or enable automated behavior of machines or other devices. Examples of applications for MTC devices include smart metering, inventory monitoring, water level monitoring, equipment monitoring, healthcare monitoring, wildlife monitoring, weather and geological event monitoring, fleet management and tracking, remote security sensing, physical access control, and transaction-based business charging.

[0071] Some UE 115s can be configured to operate in a power-saving mode, such as half-duplex communication (e.g., a mode that supports unidirectional communication via transmission or reception, but not both simultaneously). In some examples, half-duplex communication can be performed at a reduced peak rate. Other power-saving techniques for UE 115s include entering a power-saving deep sleep mode when not engaged in active communication, operating on limited bandwidth (e.g., according to narrowband communication), or a combination of these techniques. For example, some UE 115s can be configured to operate using a narrowband protocol type associated with a defined portion or range (e.g., a set of subcarriers or resource blocks (RBs)) within a carrier, within a carrier's guard band, or outside a carrier.

[0072] Wireless communication system 100 can be configured to support ultra-reliable communication or low-latency communication, or various combinations thereof. For example, wireless communication system 100 can be configured to support ultra-reliable low-latency communication (URLLC) or mission-critical communication. UE 115 can be designed to support ultra-reliable, low-latency, or mission-critical functions (e.g., mission-critical functions). Ultra-reliable communication may include private or group communication and may be supported by one or more mission-critical services, such as Mission-Critical Push-to-Talk (MCPTT), Mission-Critical Video (MCVideo), or Mission-Critical Data (MCData). Support for mission-critical functions may include service prioritization, and mission-critical services may be used for public safety or general business applications. The terms ultra-reliable, low-latency, mission-critical, and ultra-reliable low-latency are used interchangeably herein.

[0073] In some examples, UE 115 is also able to communicate directly with other UE 115 via device-to-device (D2D) communication link 135 (e.g., using peer-to-peer (P2P) or D2D protocols). One or more UE 115s utilizing D2D communication can be within the geographic coverage area 110 of base station 105. Other UE 115s in the group may be outside the geographic coverage area 110 of base station 105 or may not be able to receive transmissions from base station 105 for other reasons. In some examples, the group of UE 115s communicating via D2D communication can utilize a one-to-many (1:M) system, where each UE 115 transmits to every other UE 115 in the group. In some examples, base station 105 facilitates resource scheduling for D2D communication. In other cases, D2D communication is performed between UE 115s without involving base station 105.

[0074] In some systems, the D2D communication link 135 may be an example of a communication channel (such as a sidelink communication channel) between vehicles (e.g., UE 115). In some examples, vehicles may communicate using vehicle-to-everything (V2X) communication, vehicle-to-vehicle (V2V) communication, or some combination of these communications. Vehicles may signal information related to traffic conditions, signaling, weather, safety, emergencies, or any other information related to the V2X system. In some examples, vehicles in a V2X system may communicate with roadside infrastructure such as roadside units, or with the network via one or more network nodes (e.g., base station 105) using vehicle-to-network (V2N) communication, or both.

[0075] Core network 130 can provide user authentication, access authorization, tracking, Internet Protocol (IP) connectivity, and other access, routing, or mobility functions. Core network 130 can be an evolved packet core (EPC) or a 5G core (5GC), which may include at least one control plane entity (e.g., a mobility management entity (MME), access and mobility management function (AMF)) managing access and mobility, and at least one user plane entity (e.g., a serving gateway (S-GW), packet data network (PDN) gateway (P-GW), or user plane function (UPF)) routing packets or interconnections to external networks. The control plane entity can manage non-access stratum (NAS) functions of UE 115 served by base station 105 associated with core network 130, such as mobility, authentication, and bearer management. User IP packets can be transferred through the user plane entity, which can provide IP address allocation and other functions. The user plane entity can connect to one or more network operator IP services 150. IP services 150 may include access to the Internet, one or more intranets, IP Multimedia Subsystem (IMS), or packet-switched streaming services.

[0076] Some network devices, such as base station 105, may include sub-components such as access network entity 140, which may be an example of an access node controller (ANC). Each access network entity 140 may communicate with UE 115 through one or more other access network transport entities 145, which may be referred to as a radio head, smart radio head, or transmit / receive point (TRP). Each access network transport entity 145 may include one or more antenna panels. In some configurations, the various functions of each access network entity 140 or base station 105 may be distributed across various network devices (e.g., radio heads and ANCs) or combined into a single network device (e.g., base station 105).

[0077] Wireless communication system 100 can operate using one or more frequency bands, typically in the range of 300 MHz to 300 GHz. The region from 300 MHz to 3 GHz is generally referred to as the Ultra High Frequency (UHF) region or decimeter band because the wavelength range is from approximately 1 decimeter to 1 meter. UHF waves may be blocked or deflected by buildings and environmental features, but these waves can penetrate buildings sufficiently to provide service to UE 115 located indoors via macrocells. Compared to transmissions using smaller frequencies and longer waves in the High Frequency (HF) or Very High Frequency (VHF) portions of the spectrum below 300 MHz, UHF wave transmission can be associated with smaller antennas and shorter ranges (e.g., less than 100 km).

[0078] The wireless communication system 100 can also operate in the ultra-high frequency (SHF) region, also known as the centimeter band, using a frequency band from 3 GHz to 30 GHz, or in the extremely high frequency (EHF) spectrum region (e.g., from 30 GHz to 300 GHz), also known as the millimeter band. In some examples, the wireless communication system 100 can support millimeter-wave (mmW) communication between the UE 115 and the base station 105, and the EHF antennas of the corresponding devices can be smaller and more closely spaced than UHF antennas. In some examples, this can facilitate the use of antenna arrays within the device. However, compared to SHF or UHF transmissions, EHF transmissions may suffer from greater atmospheric attenuation and a shorter range. The techniques disclosed herein can be adopted across transmissions using one or more different frequency regions, and the designated use of frequency bands across these frequency regions can vary by country or regulatory body.

[0079] Wireless communication system 100 can utilize licensed and unlicensed radio spectrum bands. For example, wireless communication system 100 can employ Licensed Assisted Access (LAA), LTE-Unlicensed (LTE-U) radio access technology, or NR technology in unlicensed bands such as the 5 GHz Industrial, Scientific, and Medical (ISM) band. When operating in unlicensed radio bands, devices such as base station 105 and UE 115 can employ carrier awareness for collision detection and avoidance. In some examples, operation in unlicensed bands can be based on carrier aggregation configurations (e.g., LAA) that combine component carriers operating in licensed bands. Operation in unlicensed spectrum can include downlink transmission, uplink transmission, P2P transmission, or D2D transmission, etc.

[0080] Base station 105 or UE 115 may be equipped with multiple antennas that can be used to employ techniques such as transmit diversity, receive diversity, multiple-input multiple-output (MIMO) communication, or beamforming. The antennas of base station 105 or UE 115 may be located within one or more antenna arrays or antenna panels that can support MIMO operation or transmit or receive beamforming. For example, one or more base station antennas or antenna arrays may be juxtaposed at an antenna assembly (such as an antenna tower). In some examples, the antennas or antenna arrays associated with base station 105 may be located in different geographical locations. Base station 105 may have an antenna array with a number of rows and columns of antenna ports that base station 105 can use to support beamforming for communication with UE 115. Similarly, UE 115 may have one or more antenna arrays that can support various MIMO or beamforming operations. Additionally or alternatively, antenna panels may support radio frequency beamforming of signals transmitted via antenna ports.

[0081] Base station 105 or UE 115 can use MIMO communication to utilize multipath signal propagation and improve spectral efficiency by transmitting or receiving multiple signals via different spatial layers. This technique can be called spatial multiplexing. For example, multiple signals can be transmitted by a transmitting device via different antennas or different combinations of antennas. Similarly, a receiving device can receive multiple signals 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 bits associated with the same data stream (e.g., the same codeword) or different data streams (e.g., different codewords). Different spatial layers can be associated with different antenna ports used for channel measurement and reporting. MIMO techniques include single-user MIMO (SU-MIMO) and multi-user MIMO (MU-MIMO). In single-user MIMO, multiple spatial layers are transmitted to the same receiving device, while in multi-user MIMO, multiple spatial layers are transmitted to multiple devices.

[0082] Beamforming (also known as spatial filtering, directional transmission, or directional reception) is a signal processing technique that can be used at a transmitting or receiving device (e.g., base station 105, UE 115) to shape or manipulate an antenna beam (e.g., a transmit beam, a receive beam) along a spatial path between the transmitting and receiving devices. Beamforming can be achieved by combining signals transmitted via antenna elements of an antenna array such that some signals propagating in a particular direction relative to the antenna array experience constructive interference while others experience destructive interference. Adjustments to the signals transmitted via antenna elements can include the transmitting or receiving device applying amplitude offset, phase offset, or both to the signals carried via the antenna elements associated with that device. The adjustments associated with each antenna element can be defined by a beamforming weight set associated with a particular direction (e.g., relative to the antenna array of the transmitting or receiving device, or relative to some other direction).

[0083] Base station 105 or UE 115 may use beam scanning technology as part of beamforming operations. For example, base station 105 may use multiple antennas or antenna arrays (e.g., antenna panels) to perform beamforming operations for directional communication with UE 115. Some signals (e.g., synchronization signals, reference signals, beam selection signals, or other control signals) may be transmitted multiple times by base station 105 in different directions. For example, base station 105 may transmit signals according to different beamforming weight sets associated with different transmission directions. Transmissions in different beam directions may be used (e.g., by a transmitting device such as base station 105, or by a receiving device such as UE 115) to identify beam directions for subsequent transmission or reception by base station 105.

[0084] The wireless communication system 100 can be a packet-based network operating according to a layered protocol stack. In the user plane, communication at the bearer or Packet Data Convergence Protocol (PDCP) layer can be IP-based. The Radio Link Control (RLC) layer can perform packet segmentation and reassembly for communication on logical channels. The Media Access Control (MAC) layer can perform priority processing and multiplexing logical channels into transport channels. The MAC layer can also use error detection techniques, error correction techniques, or both to support retransmissions at the MAC layer to improve link efficiency. In the control plane, the Radio Resource Control (RRC) protocol layer can provide the establishment, configuration, and maintenance of RRC connections between the UE 115 and the base station 105 or core network 130 that supports user plane data radio bearers. At the physical layer, transport channels can be mapped to physical channels.

[0085] UE 115 and base station 105 can support data retransmission to improve the likelihood of successful data reception. Hybrid Automatic Repeat Request (HARQ) feedback is a technique used to improve the likelihood of correct data reception on communication link 125. HARQ can include a combination of error detection (e.g., using Cyclic Redundancy Check (CRC)), forward error correction (FEC), and retransmission (e.g., Automatic Repeat Request (ARQ)). HARQ can improve MAC layer throughput under poor radio conditions (e.g., low signal-to-noise ratio conditions). In some examples, the device can support HARQ feedback within the same time slot, where the device can provide HARQ feedback in a specific time slot for data received in a previous symbol within that time slot. In other cases, the device can provide HARQ feedback in subsequent time slots or according to some other time interval.

[0086] Machine learning, especially deep machine learning, has become a popular tool for promoting more efficient communication in wireless communication systems. For example, among other benefits, a machine learning model deployed at UE 115 can enable UE 115 to make decisions or perform actions (e.g., using prediction or regression or other objectives) without additional signaling from base station 105 (e.g., UE 115 can make inferences about the action to be performed based on inputs or detected events). Additionally, the machine learning model can enable UE 115 to prepare transmissions for more efficient communication (e.g., using classification or compression or other objectives). Before the machine learning model can be deployed on the device, it can be prepared and trained (e.g., using a dataset).

[0087] Machine learning models using neural networks can be deployed on devices (e.g., UE 115) for different applications (e.g., prediction, classification, compression, regression, or other objectives). For example, a machine learning model may include one or more parameters (e.g., a prepared dataset) that, when recognized by the device, enable or support the corresponding application at the device. To build and further train the machine learning model, the network device can collect different datasets to identify the meaning or impact of the datasets on different applications. That is, a machine learning model can be trained on one or more datasets using a neural network, and when real-world data is input into the machine learning model, the model can generate output based on the dataset.

[0088] Additionally, UE 115 can perform several channel condition measurements as part of its communication with base station 105. For example, measurements may include measurements of the channel and interference at each antenna port (e.g., channel state feedback), power measurements from the serving cell and neighboring cells, RAT-to-Reach measurements (e.g., from a WiFi network), sensor measurements, etc. These measurements can generate a considerable amount of data to be transmitted to base station 105 to assist base station 105 in network management. In some cases, UE 115 may employ machine learning models or neural networks to compress the considerable amount of data to be transmitted.

[0089] In some examples, the quantities measured by UE 115 (e.g., in a 5G network) may depend on multiple parameters. For example, multiple parameters that can affect the measurement may include antenna design and placement measurements at UE 115 and their time-varying congestion, environmental parameters (e.g., the location, shadowing, presence, and movement of reflectors near UE 115 and / or base station 105, where reflector placement can also cause inter-tap correlations such as wide-path resolution / beam splitting into many paths), cell load (e.g., causing handover), movement of the UE 115 in question (e.g., a change in its orientation), etc. To account for these types of varying parameters that may affect the measurement, UE 115 may use or be implemented by a neural network, where the neural network learns the dependence of the measurement on individual parameters, isolates those measurements through various layers, and compresses the measurement while reducing (e.g., minimizing) compression loss. For example, UE 115 may use a neural network to compress measurements to reduce the size of the transmission carrying the measurement. However, further techniques are needed to enhance the compression techniques for measurement data using neural networks.

[0090] When UE 115 compresses data in transmission before sending it to base station 105, wireless communication system 100 can support adding additional layers to the compression operation. For example, UE 115 can measure one or more channel conditions and then use a neural network to compress the measurement data. In some examples, UE 115 can quantize the output of the neural network to provide more transmittable data. Subsequently, after neural network compression, UE 115 can add additional layers to further compress the data, where UE 115 can use differential coding, entropy coding, or both to further compress and encode the measurement data before sending it to the base station. Differential coding can be an example of encoding the difference between two measurements without encoding the absolute value of the measurements. For example, differential coding can use a previous value, an initial value, a reconstructed value, an initial reconstructed value, or a combination thereof to indicate the differential value. When decoding the encoded, quantized, and compressed transmission from UE 115, base station 105 can then use differential decoding, entropy decoding, or both. In some examples, base station 105 may send indications of differential coding and entropy coding (e.g., parameters for each coding) for use by UE 115.

[0091] Figure 2 An example of a wireless communication system 200 supporting coding techniques for neural network architectures according to various aspects of this disclosure is illustrated. In some examples, the wireless communication system 200 may implement aspects of the wireless communication system 100, or may be implemented by aspects of the wireless communication system 100. For example, the wireless communication system 200 may include a base station 105-a and a UE 115-a, which may be respectively as referenced Figure 1 Examples of corresponding base station 105 and UE 115 are described below. Additionally, base station 105-a and UE 115-a can communicate on resources of carrier 205 (e.g., for downlink communication) and carrier 210 (e.g., for uplink communication). Although shown as separate carriers, carrier 205 and carrier 210 may include the same or different resources (e.g., time and frequency resources) for corresponding transmissions.

[0092] As described herein, using the properties of the quantity being compressed (e.g., data in a dataset), UE 115-a can progressively extract and compress each feature (e.g., dimension) affecting that quantity using a neural network. For example, UE 115-a can identify the feature (e.g., dimension) to be compressed. In some examples, UE 115-a can perform one type of operation in that dimension and can perform a common operation in other dimensions. For example, UE 115-a can subsequently use a fully connected layer on the first dimension and can use convolutions (e.g., pointwise convolutions) on other dimensions. UE 115-a can then perform extraction with additional stacked layers. For example, additional stacked layers can include convolutional layers, fully connected layers, or other layers with or without activation (e.g., Residual Neural Network (ResNet) layers). After extraction, the neural network employed on UE 115-a can compress the feature. For example, the neural network can use convolutional or fully connected layers or another type of layer for this compression. UE 115-a can repeat this process for subsequent features. After each feature has been extracted and compressed, UE 115-a can use one or more additional compression layers (e.g., convolutional layers, fully connected layers, or another type of layer). Using the techniques described herein, UE 115-a can then perform additional coding for the final compression, such as differential coding, entropy coding, or both.

[0093] Differential coding can involve encoding the difference between two measurements, rather than the absolute value of the measurements. For example, a first signal power measurement taken at a first time can be compared to a second signal power measurement taken at a second time. In differential coding, the difference between the first and second signal power measurements can be transmitted, rather than the absolute value of the second signal power measurement. Because some measurements do not change significantly over time, some differential data may use fewer bits than the absolute value of the data. Additionally, entropy coding can involve compressing a certain number of coded bits to reduce the number of bits transmitted. In some examples, entropy coding schemes can use symbols with a number of bits that are inversely proportional to the probability of the symbol occurring. For example, a more likely symbol can be encoded using fewer bits than a less likely symbol.

[0094] In some examples, UE 115-a may perform differential coding, entropy coding, or both to further compress the data after a quantization step associated with neural network processing. Additionally or alternatively, differential coding may occur at the input stage of the compression operation, where bits representing the differences between measurements are initially encoded and then fed into the neural network for processing. When decoding the compressed, encoded transmission from UE 115-a, base station 105-a can then use differential decoding, entropy decoding, or both.

[0095] Before performing these compression and encoding operations, base station 105-a may instruct UE 115-a to perform different compression and encoding operations. For example, base station 105-a may send an indication of encoding parameter 215 to UE 115-a (e.g., via carrier 205). In some examples, encoding parameter 215 may include differential encoder parameter 220, entropy encoder parameter 225, or both. One or more encoder parameters can enable and instruct UE 115-a to use the corresponding encoder when compressing and encoding data (e.g., measurement data) before sending data to base station 105-a. For example, UE 115-a may detect or determine the dataset 230 to be compressed and encoded based on encoding parameter 215 (e.g., including differential encoder parameter 220, entropy encoder parameter 225, or both). UE 115-a can then transmit the compressed and encoded dataset 235 to base station 105-a (e.g., via carrier 210) based on differential encoder operation (e.g., according to differential encoder parameter 220), entropy encoder operation (e.g., according to entropy encoder parameter 225), or both. Base station 105-a can then decode the compressed and encoded dataset 235 based on differential decoder operation, entropy decoder operation, or both.

[0096] In some examples, encoding parameter 215 may instruct UE 115-a how to perform the corresponding encoder operation. For example, based on differential encoder parameter 220, base station 105-a may instruct UE 115-a which previous values ​​will be used to perform differential encoder operation. In some implementations, UE 115-a may use the initial value of the data (e.g., at an intra-coded frame (I-frame)) to determine the differential value of the data at a subsequent time instance (e.g., at a prediction frame (P-frame)). Additionally or alternatively, UE 115-a may use previously reconstructed values ​​of the data to determine the differential value of the data at a subsequent time instance. In some examples, UE 115-a may use the initial reconstructed values ​​of the data to determine the differential value of the data at a subsequent time instance. Additionally or alternatively, base station 105-a may instruct UE 115-a how to perform entropy coding operation via entropy encoder parameter 225. For example, entropy encoder parameter 225 can indicate the type of entropy encoder operation to be performed (e.g., Huffman coding, arithmetic coding, embedded zero-tree wavelet (EZW) coding, Lenper-Ziff (LZ) entropy coding, etc.) and additional parameters to enable entropy encoder operation.

[0097] As described herein, UE 115-a may perform differential encoder operation, entropy encoder operation, or both at the output of a single encoder. For example, UE 115-a may first encode dataset 230 using a single encoder (e.g., an additional encoder in addition to a differential encoder and an entropy encoder) to compress and encode dataset 230. In some examples, after encoding dataset 230 using a single encoder, UE 115-a may quantize the encoded dataset (e.g., an encoded and compressed dataset). Quantization of the encoded dataset may include storing or transforming values ​​in the encoded dataset with a bit width lower than that of a floating-point prediction (e.g., floating-point values ​​are converted to bits so that values ​​in the encoded dataset are represented by integers instead of floating-point values). Once the encoded dataset has been quantized (e.g., the dataset is now encoded, quantized, and compressed), UE 115-a may then perform differential encoder operation, entropy encoder operation, or both before sending a further encoded, quantized, and compressed dataset to base station 105-a. Additionally or alternatively, UE 115-a may perform differential encoder operations at the input stage to single-encode, and perform the remaining operations described herein.

[0098] In some examples, UE 115-a can determine when to perform a differential encoder operation, an entropy encoder operation, or both (e.g., at the input or output phase) associated with a single encoder operation based on a corresponding indication parameter for each encoder operation. For example, differential encoder parameter 220 can indicate when UE 115-a performs a differential encoder operation (e.g., at the input or output phase of a single encoder operation), and entropy encoder parameter 225 can indicate when UE 115-a performs an entropy encoder operation (e.g., at the output phase of a single encoder operation). Additionally or alternatively, UE 115-a can be pre-configured for when to perform these encoder operations.

[0099] Despite Figure 2 In the example, UE 115-a is shown performing different compression and encoding operations; however, any encoding device (e.g., base station 105, TRP, another type of UE 115, etc.) can perform the techniques described herein to compress and encode the dataset before sending it to an additional device. Additionally, when base station 105-a is shown receiving and decoding compressed and encoded dataset 235 (e.g., encoded, quantized, and compressed dataset), any decoding device (e.g., UE 115, TRP, another type of base station 105, etc.) can perform the techniques described herein to decode the compressed and encoded dataset 235.

[0100] Figure 3A and 3BExamples of compression programs 300 and 301 supporting encoding techniques for neural network architectures according to various aspects of this disclosure are illustrated. In some examples, compression programs 300 and 301 may be implemented by aspects of wireless communication system 100, wireless communication system 200, or both, or may be implemented by aspects of wireless communication system 100, wireless communication system 200, or both. For example, an encoding device (e.g., UE 115 or an additional encoding device) may be configured to perform one or more operations on samples (e.g., data) received via one or more antennas of the encoding device to compress the samples. Subsequently, a decoding device (e.g., base station 105 or an additional decoding device) may be configured to decode the compressed samples to determine information such as channel state feedback.

[0101] In some examples, the encoding device can identify the features to be compressed. For example, the encoding device can perform a first type of operation in a first dimension associated with the features to be compressed. The encoding device can perform a second type of operation in other dimensions (e.g., in all other dimensions). For example, the encoding device can perform a fully connected operation in the first dimension and convolutions (e.g., pointwise convolutions) in other dimensions. The identified features to be compressed can be the input 305 (e.g., h(t) or h) for the compression procedure 300. in This is part of the compression process 300. The different operations shown for the compression process 300 may include multiple neural network layers and / or operations. The neural networks of the encoding and decoding devices can be formed by cascading one or more of the operations shown.

[0102] The encoding device may perform spatial domain feature extraction 310 on the data (e.g., input 305). Subsequently, the encoding device may perform tapped domain feature extraction 315 on the data. In some examples, the encoding device may perform tapped domain feature extraction before performing spatial domain feature extraction. Additionally, the extraction operation may include multiple operations. For example, multiple operations may include one or more convolutional operations, one or more fully connected operations, etc., which may or may not be activated. In some examples, the extraction operation may include one or more ResNet operations.

[0103] After performing spatial domain feature extraction 310 and tapped domain feature extraction 315 (e.g., extraction operations), the encoding device can compress one or more extracted features using feature compression 320. In some examples, feature compression 320 (e.g., compression operation) may include one or more operations, such as one or more convolutional operations, one or more fully connected operations, etc. After compression, the bit count of the output may be less than the bit count of the input.

[0104] Subsequently, the encoding device can perform quantization 325 (e.g., a quantization operation). In some examples, the encoding device can perform quantization 325 after the output of the flattening and compression operation and / or perform a fully connected operation after the flattened output. Using spatial domain feature extraction 310, tap domain feature extraction 315, and feature compression 320 can represent performing or using a single encoder at the encoding device. Then, the output of the single encoder can be quantized 325 before the encoding device sends the encoded data to the decoding device.

[0105] Accordingly, the decoding device can then perform feature decompression 330 based on the received encoded data. Additionally, the decoding device can perform tap domain feature reconstruction 335 and spatial domain feature reconstruction 340 to produce an output 345 corresponding to the input 305 (e.g., h...). out Or h(t)). In some examples, the decoding device may perform spatial feature reconstruction 340 before performing tap domain feature reconstruction 335. After the reconstruction operation, the decoding device may output the reconstructed version via output 345 based on the input 305 from the encoding device. Using feature decompression 330, tap domain feature reconstruction 335, and spatial feature reconstruction 340 can represent performing or using a single decoder at the decoding device.

[0106] In some examples, the decoding device may perform operations in the reverse order of those performed by the encoding device. For example, if the encoding device follows operations (A, B, C, D), the decoding device may follow the inverse operations (D, C, B, A). Additionally, the decoding device may perform operations that are completely symmetric to those performed by the encoding device. This use of symmetric operations can reduce the number of bits required for neural network configuration at the encoding device. Additionally or alternatively, the decoding device may perform additional operations (e.g., convolution operations, fully connected operations, ResNet operations, etc.) in addition to those performed by the encoding device. That is, the decoding device may perform operations that are asymmetric to those performed by the encoding device.

[0107] Based on an encoding device that encodes a dataset using a neural network for uplink communication, the encoding device (e.g., UE 115) can transmit measurements (e.g., channel state feedback) with a reduced payload. This reduced payload saves network resources that might otherwise be used to transmit the complete dataset sampled by the encoding device. (See references herein.) Figure 2 As described, the encoding device can use an additional layer to perform additional encoding operations to further compress the data to be sent to the decoding device.

[0108] For example, as shown in reference compression procedure 301, the encoding device may apply or use a differential encoder 360, an entropy encoder 365, or both on the output of a single encoder to further compress the feedback 370 to be sent to the decoding device. Feedback 370 may be an example of information transmitted between the transmitting and receiving devices via an air interface (or some other medium). Additionally, in performing reference... Figure 3A Prior to the operation of the single-pass decoder described in compression procedure 300, the decoding device may also use an entropy decoder 375, a differential decoder 380, or both. Additionally or alternatively, although not shown, the encoding device may apply or use a differential encoder 360 at the input stage of the single-pass encoder, and the decoding device may apply or use a differential decoder 380 at the output of the single-pass encoder. In some examples, compression procedure 301 may be referred to as a time-domain compression operation or procedure.

[0109] like Figure 3B As shown in the example, the encoding device can acquire input 305 and pass it through encoder neural network 350. In some examples, encoder neural network 350 may correspond to, as in reference [reference] Figure 3A The described spatial feature extraction 310, tapped domain feature extraction 315, and feature compression 320 (e.g., a single-pass encoder). After the input 305 has passed through the encoder neural network 350, the encoding device can use the quantizer 355 to perform a quantization operation (e.g., reference). Figure 3A The quantization is described in section 325. Subsequently, the encoding device can then apply differential encoder 360, entropy encoder 365, or both, or apply differential encoder 360, entropy encoder 365, or both to the quantized output of encoder neural network 350. In some examples, differential encoder 360 and entropy encoder 365 can be as described in reference... Figure 2 It is performed as described. The encoding device can then send feedback 370 to the decoding device, where feedback 370 represents the encoded, quantized, and compressed dataset.

[0110] The decoding device can receive feedback 370 and can use or apply entropy decoder 375, differential decoder 380, or both to partially decode and decompress the encoded, quantized, and compressed dataset received via feedback 370. After using or applying entropy decoder 375, differential decoder 380, or both, the decoding device can then pass the partially decoded and decompressed dataset via decoder neural network 385 to generate output 345. In some examples, decoder neural network 385 may include, as in reference [reference missing] Figure 3A The described feature decompression 330, tap domain feature reconstruction 335, and spatial domain feature reconstruction 340 (e.g., and any additional layers).

[0111] Figure 4 An example of a compression configuration 400 supporting coding techniques for neural network architectures according to various aspects of this disclosure is illustrated. In some examples, compression configuration 400 may be implemented by or by aspects of wireless communication system 100, wireless communication system 200, or both. For example, an encoding device (e.g., UE 115 or an additional encoding device) may be configured to perform one or more operations on samples (e.g., data) received via one or more antennas of the encoding device to compress the samples. Subsequently, a decoding device (e.g., base station 105 or an additional decoding device) may be configured to decode the compressed samples to determine information such as channel state feedback. In some examples, compression configuration 400 may represent an example of a channel state feedback compression operation or procedure with entropy coding.

[0112] The encoding device can identify input 405 for performing compression configuration 400. For example, input 405 may include a dataset to be compressed and encoded (e.g., multiple measurements, such as channel state feedback measurements). In some examples, the encoding device may receive samples from antennas, where input 405 represents a 64x64-dimensional dataset received from its antennas based on the number of antennas, the number of samples per antenna, and tap features. A 64x64-dimensional dataset can represent a non-limiting example of input 405. The encoding device may pass input 405 to or through a first convolutional layer 410. For example, the first convolutional layer 410 may represent an initial layer for spatial feature extraction and short temporal (tap) feature extraction using one-dimensional convolutions (e.g., Conv1D). The first convolutional layer 410 (e.g., an additional convolutional layer) may be a fully connected layer (e.g., in the antenna) and a simple convolution with a small kernel size (e.g., 3) in the tap dimension (to extract short tap features). The output of this 64xW one-dimensional convolution operation can be a Wx64 matrix.

[0113] Following the first convolutional layer 410, the encoding device can perform one or more ResNet operations using one or more corresponding ResNet blocks (such as ResNet block 415 and ResNet block 420). One or more ResNet operations can also refine spatial and / or temporal features. In some examples, a ResNet operation may include multiple operations associated with a feature. For example, a ResNet operation may include multiple (e.g., three) one-dimensional convolutional operations, skip connections (e.g., between the input and output of the ResNet to avoid applying one-dimensional convolutional operations), summation operations through paths of multiple one-dimensional convolutional operations and paths through skip connections, or append operations. In some examples, multiple one-dimensional convolutional operations may include: a Wx256 convolutional operation with a kernel size of 3, whose output is fed into a batch normalization (BN) layer, followed by a corrected linear unit (ReLU) activation (e.g., LeakyReLU activation) that produces a 256x64-dimensional output dataset; a 256x512 convolutional operation with a kernel size of 3, whose output is fed into a BN layer, followed by a ReLU activation that produces a 512x64-dimensional output dataset; and a 512xW convolutional operation with a kernel size of 3, whose output is a BN dataset of dimension Wx64. The output from one or more ResNet operations may be a Wx64 matrix.

[0114] The encoding device can use additional convolutional layers 425 to perform WxV convolution operations on the outputs of one or more ResNet operations. WxV convolution operations can include pointwise (e.g., tapwise) convolution operations. WxV convolution operations can compress spatial features into a reduced dimension for each tap. A WxV convolution operation can have W features as input and V features as output. The output of a WxV convolution operation can be a Vx64 matrix.

[0115] The encoding device can perform a flattening operation 430 to flatten the Vx64 matrix into a 64V element vector. The encoding device can use a fully connected layer 435 to perform a 64VxM fully connected operation to further compress the spatiotemporal feature dataset into a low-dimensional vector of size M for over-the-air transmission to the decoding device. The encoding device can perform quantization 440 before over-the-air transmission of the low-dimensional vector of size M to map the transmitted samples to discrete values ​​of the low-dimensional vector of size M. After quantization 440, the encoding device can use or apply an entropy encoder 445 to further compress the output of quantization 440.

[0116] The decoding device can receive entropy-encoded data and can use or apply an entropy decoder 450 to decode the encoded transmission. Additionally, the decoding device can use a fully connected layer 455 to perform an Mx64V fully connected operation to decompress a low-dimensional vector of size M into a spatiotemporal feature dataset. The decoding device can perform a shaping operation 460 to shape a 64V element vector into a two-dimensional Vx64 matrix. The decoding device can use a convolutional layer 465 to perform a VxW convolution operation (with a kernel size of 1) on the output from the shaping operation 460. The VxW convolution operation can include pointwise (e.g., tapwise) convolution operations. In some examples, the VxW convolution operation can decompress spatial features from a reduced dimension for each tap. The VxW convolution operation can have an input of V features and an output of W features. The output from the VxW convolution operation can be a Wx64 matrix.

[0117] The decoding device can then use one or more corresponding ResNet blocks (e.g., ResNet block 470 and ResNet block 475) to perform one or more ResNet operations. One or more ResNet operations can also decompress spatial features, temporal features, or both. In some examples, ResNet operations may include multiple (e.g., three) one-dimensional convolutional operations, skip connections (e.g., to avoid applying one-dimensional convolutional operations), summation operations through paths of multiple convolutional operations, and append operations through paths of skipped connections. The output from one or more ResNet operations may be a Wx64 matrix.

[0118] The decoding device can then use convolutional layer 480 to perform spatial and temporal feature reconstruction. In some examples, spatial and temporal feature reconstruction can be achieved using a one-dimensional convolution operation that is fully connected in the spatial dimension (to reconstruct spatial features) and a simple convolution with a small kernel size (e.g., 3) in the tap dimension (to reconstruct short-tap features). The output from the 64xW convolution operation can be a 64x64 matrix and can be represented by output 485. In some examples, the values ​​of M, W, and / or V can be configurable to adjust feature weights, payload size, etc. Additionally, although in Figure 4 The examples described in this paper correspond to operations for 64x64 dimensional datasets (e.g., 64x64 matrices), but datasets of any size can be used to perform the operations described herein.

[0119] Figure 5An example of a compression and encoding configuration 500 supporting encoding techniques for neural network architectures according to various aspects of this disclosure is illustrated. In some examples, the compression and encoding configuration 500 may be implemented by or by aspects of wireless communication system 100, wireless communication system 200, or both. For example, an encoding device (e.g., UE 115 or an additional encoding device) may be configured to perform one or more operations on input 505 (e.g., data) of samples received via one or more antennas of the encoding device to compress the samples. Subsequently, a decoding device (e.g., base station 105 or an additional decoding device) may be configured to decode the compressed samples based on input 505 to determine output 530 of information, such as channel state feedback. In some examples, the compression and encoding configuration 500 may represent a compression and encoding procedure including a differential encoder 510, an entropy encoder 515, or both at the output of an encoder (e.g., a single-shot encoder, an encoder neural network, etc.). By using or applying a differential encoder 510 at the output of the encoder, the encoding device can reduce or limit error propagation.

[0120] The encoding device can take an input 505 and can use one or more operations of the encoder (such as one or more convolutional layers, one or more ResNet blocks, one or more fully connected layers, and quantizers, as referenced). Figure 4 The input 505 is initially encoded, compressed, and quantized (as described). After quantization (e.g., at the output of the encoder), the encoding device can use or apply a differential encoder 510. See reference... Figure 2 As described, the differential encoder 510 can encode data using previous values ​​of the data. In some examples, the decoding device or network device can instruct the encoding device which previous values ​​to use for the differential encoder 510 (e.g., via one or more parameters used for differential encoding operations). For example, the encoding device can use the initial value of the data (e.g., at an I-frame) to determine the differential value of the data at a later time instance (e.g., at a P-frame), use the previously reconstructed value of the data to determine the differential value of the data at a later time instance, use the initial reconstructed value of the data to determine the differential value of the data at a later time instance, or a combination thereof.

[0121] For example, when using the initial value of the data to determine the difference value of that data at a later time instance, the encoding device can calculate the P subframe (e.g., P frame) relative to the I frame. The given difference value for the time instance (N-1) can be given by Equation 1.

[0122] (x N-1 -x0)(1)

[0123] x0 can represent the initial value of the data (e.g., determined at the I-frame), and x N-1 This can represent a given value of data at time N-1 (e.g., determined at a P-frame). The encoding device can divide the difference value of the data given by Equation 1 by... Where E represents the error propagation value, and Δ represents the given value of the data (e.g., x). N-1 The difference between the original value and the initial value of the data (e.g., x0). Divide the difference by... After that, the encoding device can have E left. P / D P Value, where E P D represents the difference value at a given P-frame. P Indicates the position at a given P-frame

[0124] On the decoding device side, the decoding device can receive signals from E. P / D P The given data representation. The decoding device can then multiply the received data representation (e.g., an encoded version of the data) by... This generates a reconstructed version of Equation 1, as given by Equation 2.

[0125]

[0126] Then, the decoding device can reconstruct the values ​​from the initial values ​​of the data determined by the decoding device (e.g., at the I-frame). Add this to Equation 2 to determine and obtain the reconstructed value of the data at time N-1 (e.g., By using a differential encoder 510 based on initial values ​​of the data (e.g., x0), error propagation of the differential encoder 510 can be based solely on reconstructed values ​​(e.g., x0) from the initial values ​​of the data. ).

[0127] Additionally or alternatively, when using or applying the differential encoder 510, the encoding device may use a P-subframe structure and reconstruction based on the reconstructed channel (e.g., based on one or more parameters signaled to the encoding device of the differential encoder 510). Using the P-subframe structure and reconstruction based on the reconstructed channel can reduce or prevent error propagation. In some examples, the encoding device may determine a reconstructed value for a given data value. For example, the encoding device may determine a value divided by... The initial value of the data (e.g., x0), which is then determined by E I / D I Given. On the decoding device side, E I / D I Can be multiplied To determine the reconstructed values ​​of the initial values ​​of the data, where the reconstructed values ​​are determined by... Given. The encoding device can determine [the target] by performing actions similar to those of the decoding device. Alternatively, an indication of the reconstructed value can be received from the decoding device.

[0128] Subsequently, for a given P-frame following an I-frame (e.g., P-frame 1, P-frame 2, etc.), the encoding device can use the reconstructed value of the previous value to encode the difference value of the value at the given P-frame. For example, for P-frame N-1 (e.g., at time N-1), the encoding device can determine the difference value of the data at time N-1 based on Equation 3.

[0129]

[0130] The encoding device can then perform a similar operation as described above (e.g., dividing the difference value of the data given by Equation 3 by...). Decoding devices can receive signals from E P / D P The given data representation, and the received data representation (e.g., an encoded version of the data) can be multiplied by... To generate a refactored version of Equation 3 (e.g., The decoding device can then add reconstructed values ​​of the data determined by the decoding device from previous values ​​(e.g., ), to determine and obtain the reconstructed value of the data at time N-1 (e.g., Additionally or alternatively, it may be based on Instead of using the immediate preceding reconstructed values ​​of the data to construct P-frames.

[0131] Accordingly, the decoding device can receive the encoded, quantized, and compressed dataset from the encoding device, and can partially decode the encoded, quantized, and compressed dataset using the entropy decoder 520 and the differential decoder 525. Additionally, the decoding device can use, as referenced... Figure 3A The described one or more fully connected layers, one or more convolutional layers, and one or more ResNet blocks are used to determine and obtain the output 530 corresponding to the input 505.

[0132] Figure 6An example of a compression and encoding configuration 600 supporting encoding techniques for neural network architectures according to various aspects of this disclosure is illustrated. In some examples, the compression and encoding configuration 600 may be implemented by aspects of wireless communication system 100, wireless communication system 200, or both, or may be implemented by aspects of wireless communication system 100, wireless communication system 200, or both. For example, an encoding device (e.g., UE 115 or an additional encoding device) may be configured to perform one or more operations on input 605 (e.g., data) of samples received via one or more antennas of the encoding device to compress the samples. Subsequently, a decoding device (e.g., base station 105 or an additional decoding device) may be configured to decode the compressed samples based on input 605 to determine output 640 information such as channel state feedback. In some examples, the compression and encoding configuration 600 may represent a compression and encoding procedure including a differential encoder 610 at the input of an encoder (e.g., a single-pass encoder, an encoder neural network, etc.) and an entropy encoder 620 at the output of the encoder.

[0133] Using the differential encoder 610, the encoding device can use I-frames and one or more P-frames. I-frames may appear once every N subframes, and P-frames may appear for the remaining N-1 subframes. In some examples, the encoding device may use two (2) neural networks for the differential encoder 610, such as a first neural network for I-frames and a second neural network for (one or more) P-frames. The encoding device may be based on a reference... Figure 5 The described technique determines the initial value of the data at the I-frame (e.g., x0 divided by). (etc.). Subsequently, for each P-frame, the difference value can be determined based on the data's immediate preceding neighbor value. For example, for time N-1, the difference value of the data can be given by Equation 4.

[0134] (x N-1 -x N-2 (4)

[0135] For reference Figure 5 As described, the decoding device can then base its decoding on that difference value (e.g., multiplied by...). ) and the reconstructed value of the previous value (e.g., To determine the reconstructed value of the data for each time instance, in order to determine the reconstructed value of the data at time N–1 (e.g., The decoding device (e.g., and / or the encoding device) can then calculate the normalized mean square error (NMSE) based on the reconstructed value given by Equation 5.

[0136]

[0137] In addition to using or applying a differential encoder 610 at the input of the encoder, the encoding device may perform a normalization operation 615 to normalize the differentially encoded data before further encoding. Additionally, the decoding device may perform a de-normalization operation 630 before using or applying a differential decoder 635. In some examples, the decoding device may also include an entropy decoder 625 that decodes the output of the entropy encoder 620.

[0138] Figure 7 An example of a processing flow 700 supporting coding techniques for neural network architectures according to various aspects of this disclosure is illustrated. In some examples, the processing flow 700 may implement aspects of wireless communication system 100, wireless communication system 200, or both, or may be implemented by aspects of wireless communication system 100, wireless communication system 200, or both. For example, the processing flow 700 may include base station 105-b and UE 115-b, which may respectively represent as referenced in the reference... Figure 1-6 Examples of corresponding base station 105 and UE 115 are described.

[0139] In the following description of processing flow 700, the operations between UE 115-b and base station 105-b may be performed in different orders or at different times. Some operations may also be excluded from processing flow 700, or other operations may be added to processing flow 700. When UE 115-b and base station 105-b are shown performing a certain number of operations in processing flow 700, any wireless device may perform the operations shown.

[0140] At 705, UE 115-b may receive an instruction for one or more encoding operations for encoding a compressed dataset, including differential encoding operations or entropy encoding operations or both. For example, UE 115-b may receive one or more parameters corresponding to one or more encoding operations, each of the one or more parameters corresponding to a corresponding encoding operation of the one or more encoding operations, wherein the quantized and compressed dataset is encoded based on the one or more parameters.

[0141] At 710, UE 115-b can encode the dataset at UE 115-b or by a neural network employed by UE 115-b to generate a compressed dataset. At 715, UE 115-b can quantize the compressed dataset encoded by the neural network.

[0142] At 720, UE 115-b can encode the quantized and compressed dataset based on instructions received for one or more encoding operations. In some examples, UE 115-b can use entropy coding operations to encode the quantized and compressed dataset. For example, entropy coding operations may include encoding the compressed dataset using one or more symbols with lengths varying based on the probability of symbol occurrence.

[0143] Additionally or alternatively, UE 115-b may use differential coding operations to encode the quantized and compressed dataset. For example, differential coding operations may include encoding a certain amount of data based on previous values ​​of a certain amount of data in the compressed dataset. In some examples, UE 115-b may, after quantizing the compressed dataset encoded by a neural network, determine a difference value between a first value of data in the quantized and compressed dataset at a first time instance and a second value of data at a second time instance after the first time instance, wherein the difference value is determined based on an instruction for one or more coding operations (e.g., an instruction instructing UE 115-b to use an initial value when determining the difference value), and the quantized and compressed dataset is encoded based on the difference value. Additionally or alternatively, UE 115-b may, after quantizing the compressed dataset encoded by the neural network, determine a difference between a first reconstructed value of the data in the quantized and compressed dataset at a first time instance and a second reconstructed value of the data at a second time instance after the first time instance, wherein the difference is determined based on an instruction for one or more encoding operations (e.g., an instruction for UE 115-b to use the previously reconstructed value when determining the difference), and the quantized and compressed dataset is encoded based on the difference.

[0144] In some examples, when performing differential encoding, UE 115-b can determine initial reconstruction values ​​of data at an initial time instance associated with the encoded dataset in the quantized and compressed dataset after quantizing the compressed dataset encoded by a neural network. Subsequently, UE 115-b can determine, after quantization, a difference between the additional reconstruction values ​​of the data at an additional time instance following the initial time instance and the initial reconstruction values ​​of the data, wherein the difference value is determined based on an instruction for one or more encoding operations (e.g., an instruction instructing UE 115-b to use the initial reconstruction values ​​when determining the difference value), and the quantized and compressed dataset is encoded based on this difference value.

[0145] At 725, UE 115-b may transmit an encoded, quantized, and compressed dataset (e.g., a compressed and encoded dataset after a single encoder operation) to a second device (e.g., base station 105-b) after encoding the compressed dataset based on one or more encoding operations. In some examples, the encoded, quantized, and compressed dataset may include differential values ​​of the data in the dataset, which are based on the initial value of the data, the previously reconstructed value of the data, the initial reconstructed value of the data, or a combination thereof.

[0146] At 730, base station 105-b can decode the encoded, quantized, and compressed dataset received from UE 115-b. For example, base station 105-b can first decode the encoded, quantized, and compressed dataset to generate a compressed dataset based on one or more encoding operations. Additionally, base station 105-b can then decode the compressed dataset via a neural network (e.g., at or by base station 105-b) to generate a dataset based on the decoding of the encoded, quantized, and compressed dataset.

[0147] Figure 8 A block diagram 800 of a device 805 supporting coding techniques for neural network architectures according to various aspects of this disclosure is shown. Device 805 may be an example of various aspects of UE 115 as described herein. Device 805 may include a receiver 810, a transmitter 815, and a communication manager 820. Device 805 may also include a processor. Each of these components may communicate with each other (e.g., via one or more buses).

[0148] Receiver 810 may provide components 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 coding techniques used in neural network architectures). The information may be transmitted to other components of device 805. Receiver 810 may utilize a single antenna or a collection of multiple antennas.

[0149] Transmitter 815 may provide components for transmitting signals generated by other components of device 805. For example, transmitter 815 may transmit information associated with various information channels (e.g., control channels, data channels, information channels related to coding techniques used in neural network architectures), such as packets, user data, control information, or any combination thereof. In some examples, transmitter 815 may be juxtaposed with receiver 810 in a transceiver module. Transmitter 815 may utilize a single antenna or a collection of multiple antennas.

[0150] The communication manager 820, receiver 810, transmitter 815, or various combinations thereof, or various components thereof, may be examples of components used to perform aspects of the coding techniques for neural network architectures described herein. For example, the communication manager 820, receiver 810, transmitter 815, or various combinations thereof, or components thereof, may support methods for performing one or more functions described herein.

[0151] In some examples, the communication manager 820, receiver 810, transmitter 815, or various combinations or components thereof may be implemented in hardware (e.g., in communication management circuitry). The hardware may include a processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof, configured to or otherwise supporting components for performing the functions described in this disclosure. In some examples, the processor and memory coupled to the processor may be configured to perform one or more functions described herein (e.g., by the processor executing instructions stored in memory).

[0152] Additionally or alternatively, in some examples, the communication manager 820, receiver 810, transmitter 815, or various combinations or components thereof may be implemented using processor-run code (e.g., as communication management software or firmware). If implemented using processor-run code, the functionality of the communication manager 820, receiver 810, transmitter 815, or various combinations or components thereof may be performed by a general-purpose processor, DSP, central processing unit (CPU), ASIC, FPGA, or any combination of these or other programmable logic devices (e.g., components configured or otherwise supported for performing the functions described in this disclosure).

[0153] In some examples, the communication manager 820 can be configured to use or cooperate with the receiver 810, the transmitter 815, or both, or otherwise, to perform various operations (e.g., receiving, monitoring, sending). For example, the communication manager 820 can receive information from the receiver 810, send information to the transmitter 815, or integrate with the receiver 810, the transmitter 815, or a combination of both, to receive information, send information, or perform various other operations described herein.

[0154] According to the examples disclosed herein, the communication manager 820 can support wireless communication at the UE. For example, the communication manager 820 can be configured or otherwise support components for receiving instructions for one or more encoding operations for encoding a compressed dataset, including differential coding operations or entropy coding operations, or both. The communication manager 820 can be configured or otherwise support components for encoding the dataset by a neural network to generate a compressed dataset. The communication manager 820 can be configured or otherwise support components for quantizing the compressed dataset encoded by the neural network. The communication manager 820 can be configured or otherwise support components for encoding a quantized and compressed dataset based on received instructions for one or more encoding operations. The communication manager 820 can be configured or otherwise support components for transmitting the encoded, quantized, and compressed dataset to a second device after encoding the compressed dataset based on one or more encoding operations.

[0155] By including or configuring the communication manager 820 according to the examples described herein, device 805 (e.g., a processor that controls or is otherwise coupled to receiver 810, transmitter 815, communication manager 820, or a combination thereof) can support techniques for more efficient use of communication resources. For example, by using differential encoder operation, entropy encoder operation, or both, communication manager 820 can further compress the dataset for transmission to a second device, thereby reducing signaling overhead and using fewer communication resources to carry the compressed dataset.

[0156] Figure 9 A block diagram 900 of a device 905 supporting coding techniques for neural network architectures according to various aspects of this disclosure is shown. Device 905 may be an example of aspects of device 805 or UE 115 as described herein. Device 905 may include a receiver 910, a transmitter 915, and a communication manager 920. Device 905 may also include a processor. Each of these components may communicate with each other (e.g., via one or more buses).

[0157] Receiver 910 may provide components 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 coding techniques used in neural network architectures). The information may be transmitted to other components of device 905. Receiver 910 may utilize a single antenna or a collection of multiple antennas.

[0158] Transmitter 915 may provide components for transmitting signals generated by other components of device 905. For example, transmitter 915 may transmit information associated with various information channels (e.g., control channels, data channels, information channels related to coding techniques used in neural network architectures), such as packets, user data, control information, or any combination thereof. In some examples, transmitter 915 may be co-located with receiver 910 in a transceiver module. Transmitter 915 may utilize a single antenna or a collection of multiple antennas.

[0159] Device 905 or its various components may be examples of parts for performing aspects of the encoding techniques for neural network architectures described herein. For example, communication manager 920 may include encoder instruction component 925, neural network encoder component 930, quantization component 935, encoder component 940, encoded dataset transmission component 945, or any combination thereof. Communication manager 920 may be examples of aspects of communication manager 820 as described herein. In some examples, communication manager 920 or its various components may be configured to use receiver 910, transmitter 915, or both, or otherwise cooperate with them to perform various operations (e.g., receiving, monitoring, transmitting). For example, communication manager 920 may receive information from receiver 910, transmit information to transmitter 915, or integrate with receiver 910, transmitter 915, or a combination thereof to receive information, transmit information, or perform various other operations described herein.

[0160] According to the examples disclosed herein, the communication manager 920 can support wireless communication at the UE. The encoder instruction component 925 can be configured or otherwise supported for receiving instructions for one or more encoding operations for encoding a compressed dataset, including differential coding operations or entropy coding operations, or both. The neural network encoder component 930 can be configured or otherwise supported for encoding a dataset by a neural network to generate a compressed dataset. The quantization component 935 can be configured or otherwise supported for quantizing the compressed dataset encoded by the neural network. The encoder component 940 can be configured or otherwise supported for encoding a quantized and compressed dataset based on received instructions for one or more encoding operations. The encoded dataset transmission component 945 can be configured or otherwise supported for transmitting the encoded, quantized, and compressed dataset to a second device after encoding the compressed dataset based on one or more encoding operations.

[0161] Figure 10A block diagram 1000 of a communication manager 1020 supporting encoding techniques for neural network architectures according to various aspects of this disclosure is shown. As described herein, the communication manager 1020 may be an example of a communication manager 820, a communication manager 920, or aspects thereof. The communication manager 1020 or its various components may be examples of parts for performing various aspects of the encoding techniques for neural network architectures described herein. For example, the communication manager 1020 may include an encoder instruction component 1025, a neural network encoder component 1030, a quantization component 1035, an encoder component 1040, a encoded dataset transmission component 1045, a differential encoder component 1050, an entropy encoder component 1055, or any combination thereof. Each of these components may communicate with each other directly or indirectly (e.g., via one or more buses).

[0162] Based on the examples disclosed in this article, Communication Manager 10 2 0 can support wireless communication at the UE. Encoder indication component 1025 can be configured or otherwise supported for receiving indications of one or more encoding operations for encoding a compressed dataset, including differential coding operations or entropy coding operations, or both. Neural network encoder component 1030 can be configured or otherwise supported for encoding a dataset by a neural network to generate a compressed dataset. Quantization component 1035 can be configured or otherwise supported for quantizing the compressed dataset encoded by the neural network. Encoder component 1040 can be configured or otherwise supported for encoding a quantized and compressed dataset based on received indications of one or more encoding operations. Encoded dataset transmission component 1045 can be configured or otherwise supported for transmitting the encoded, quantized, and compressed dataset to a second device after encoding the compressed dataset based on one or more encoding operations.

[0163] In some examples, in order to support receiving instructions for one or more encoding operations, the encoder instruction component 1025 may be configured or otherwise supported for receiving a component for receiving one or more parameters corresponding to one or more encoding operations, each of the one or more parameters corresponding to a corresponding encoding operation of the one or more encoding operations, wherein the quantized and compressed dataset is encoded based on one or more parameters.

[0164] In some examples, to support encoding of quantized and compressed datasets, the differential encoder component 1050 can be configured or otherwise supported for encoding quantized and compressed datasets using differential encoding operations after the datasets have been encoded using a neural network.

[0165] In some examples, to support encoding of quantized and compressed datasets, the entropy encoder component 1055 can be configured or otherwise supported as a component for encoding quantized and compressed datasets using entropy encoding operations after the datasets have been encoded using a neural network.

[0166] In some examples, to support encoding of quantized and compressed datasets, the differential encoder component 1050 may be configured or otherwise support a component for determining a difference value between a first value of data in the quantized and compressed dataset at an initial time instance and a second value of data at a second time instance after the initial time instance, after quantizing the compressed dataset encoded by the neural network, wherein the difference value is determined based on an instruction of one or more encoding operations, and the quantized and compressed dataset is encoded based on the difference value.

[0167] In some examples, to support encoding of quantized and compressed datasets, the differential encoder component 1050 may be configured or otherwise support a component for determining a difference value between a first reconstructed value of data in the quantized and compressed dataset at an initial time instance and a second reconstructed value of data at a second time instance after the initial time instance, after quantizing the compressed dataset encoded by the neural network, wherein the difference value is determined based on an instruction of one or more encoding operations, and the quantized and compressed dataset is encoded based on the difference value.

[0168] In some examples, to support encoding of quantized and compressed datasets, the differential encoder component 1050 may be configured or otherwise supported to include components for determining initial reconstructed values ​​of data in the quantized and compressed datasets at an initial time instance associated with the encoded dataset after quantization of the compressed dataset encoded by the neural network. In some examples, to support encoding of quantized and compressed datasets, the differential encoder component 1050 may be configured or otherwise supported to include components for determining, after quantization, a difference between the additional reconstructed values ​​of the data at an additional time instance following the initial time instance and the initial reconstructed values ​​of the data, wherein the difference is determined based on indications of one or more encoding operations, and the quantized and compressed datasets are encoded based on the difference values.

[0169] In some examples, differential coding involves encoding a certain amount of data based on previous values ​​of a certain amount of data in a compressed dataset.

[0170] In some examples, entropy coding operations involve encoding a compressed dataset using one or more symbols of varying lengths based on the probability of their occurrence.

[0171] Figure 11 A diagram of a system 1100 including a device 1105 supporting coding techniques for neural network architectures is shown according to various aspects of this disclosure. Device 1105 may be an example of or include components of device 805, device 905, or UE 115 as described herein. Device 1105 may wirelessly communicate with one or more base stations 105, UE 115, or any combination thereof. Device 1105 may include components for bidirectional voice and data communication, including components for transmitting and receiving communications, such as a communication manager 1120, an input / output (I / O) controller 1110, a transceiver 1115, an antenna 1125, a memory 1130, code 1135, and a processor 1140. These components may communicate electronically or be otherwise coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more buses (e.g., bus 1145).

[0172] I / O controller 1110 can manage the input and output signals of device 1105. I / O controller 1110 can also manage peripheral devices not integrated into device 1105. In some cases, I / O controller 1110 can represent a physical connection or port to an external peripheral device. In some cases, I / O controller 1110 can utilize, for example... MS- MS- Or another known operating system. Additionally or alternatively, the I / O controller 1110 may represent or interact with a modem, keyboard, mouse, touchscreen, or similar device. In some cases, the I / O controller 1110 may be implemented as part of a processor, such as processor 1140. In some cases, a user may interact with device 1105 via the I / O controller 1110 or via hardware components controlled by the I / O controller 1110.

[0173] In some cases, device 1105 may include a single antenna 1125. However, in other cases, device 1105 may have more than one antenna 1125, capable of transmitting or receiving multiple wireless transmissions simultaneously. Transceiver 1115 may communicate bidirectionally via one or more antennas 1125, wired or wireless links, as described herein. For example, transceiver 1115 may represent a wireless transceiver and may communicate bidirectionally with another wireless transceiver. Transceiver 1115 may also include a modem for modulating packets, providing modulated packets to one or more antennas 1125 for transmission, and demodulating packets received from one or more antennas 1125. As described herein, transceiver 1115 or transceiver 1115 and one or more antennas 1125 may be examples of transmitter 815, transmitter 915, receiver 810, receiver 910, or any combination thereof or components thereof.

[0174] Memory 1130 may include random access memory (RAM) and read-only memory (ROM). Memory 1130 may store computer-readable, computer-executable code 1135, including instructions that, when executed by processor 1140, cause device 1105 to perform the various functions described herein. Code 1135 may be stored in a non-transitory computer-readable medium, such as system memory or another type of memory. In some cases, code 1135 may not be directly executable by processor 1140, but may cause a computer (e.g., when compiled and run) to perform the functions described herein. In some cases, among others, memory 1130 may contain a basic I / O system (BIOS) that controls basic hardware or software operations, such as interaction with peripheral components or devices.

[0175] Processor 1140 may include intelligent hardware devices (e.g., general-purpose processors, DSPs, CPUs, microcontrollers, ASICs, FPGAs, programmable logic devices, discrete gate or transistor logic components, discrete hardware components, or any combination thereof). In some cases, processor 1140 may be configured to use a memory controller to operate a memory array. In other cases, the memory controller may be integrated into processor 1140. Processor 1140 may be configured to execute computer-readable instructions stored in memory (e.g., memory 1130) to cause device 1105 to perform various functions (e.g., functions or tasks supporting coding techniques for neural network architectures). For example, device 1105 or components of device 1105 may include processor 1140 and memory 1130 coupled to processor 1140, processor 1140 and memory 1130 being configured to perform the various functions described herein.

[0176] According to the examples disclosed herein, the communication manager 1120 can support wireless communication at the UE. For example, the communication manager 1120 can be configured or otherwise support components for receiving instructions for one or more encoding operations for encoding a compressed dataset, including differential coding operations or entropy coding operations, or both. The communication manager 1120 can be configured or otherwise support components for encoding the dataset by a neural network to generate a compressed dataset. The communication manager 1120 can be configured or otherwise support components for quantizing the compressed dataset encoded by the neural network. The communication manager 1120 can be configured or otherwise support components for encoding a quantized and compressed dataset based on received instructions for one or more encoding operations. The communication manager 1120 can be configured or otherwise support components for transmitting the encoded, quantized, and compressed dataset to a second device after encoding the compressed dataset based on one or more encoding operations.

[0177] By including or configuring the communication manager 1120 according to the examples described herein, device 1105 can support techniques for more efficient use of communication resources, improved coordination between devices, and improved utilization of processing power. For example, by using differential encoder operation, entropy encoder operation, or both, communication manager 1120 can further compress the dataset for transmission to a second device, thereby reducing signaling overhead and using fewer communication resources to carry the compressed dataset. Additionally, by receiving parameter indications for different encoder operations from the network device, communication manager 1120 can support coordination between device 1105 and the network device. Using differential encoder operation, entropy encoder operation, or both can also utilize more processing power of device 1105 instead of performing fewer encoding operations, resulting in less compressed data than without differential encoder operation and entropy encoder operation.

[0178] In some examples, the communication manager 1120 may be configured to perform various operations (e.g., receiving, monitoring, transmitting) using or otherwise cooperating with a transceiver 1115, one or more antennas 1125, or any combination thereof. 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 may be supported or performed by a processor 1140, memory 1130, code 1135, or any combination thereof. For example, code 1135 may include instructions executable by processor 1140 to cause device 1105 to perform aspects of the coding techniques for neural network architectures as described herein, or processor 1140 and memory 1130 may be otherwise configured to perform or support such operations.

[0179] Figure 12 A block diagram 1200 of a device 1205 supporting encoding techniques for neural network architectures according to various aspects of this disclosure is shown. Device 1205 may be an example of various aspects of base station 105 as described herein. Device 1205 may include receiver 1210, transmitter 1215, and communication manager 1220. Device 1205 may also include a processor. Each of these components may communicate with each other (e.g., via one or more buses).

[0180] Receiver 1210 may provide components 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 coding techniques used in neural network architectures). The information may be transmitted to other components of device 1205. Receiver 1210 may utilize a single antenna or a collection of antennas.

[0181] Transmitter 1215 may provide components for transmitting signals generated by other components of device 1205. For example, transmitter 1215 may transmit information associated with various information channels (e.g., control channels, data channels, information channels related to coding techniques used in neural network architectures), such as packets, user data, control information, or any combination thereof. In some examples, transmitter 1215 may be co-located with receiver 1210 in a transceiver module. Transmitter 1215 may utilize a single antenna or a collection of multiple antennas.

[0182] The communication manager 1220, receiver 1210, transmitter 1215, or various combinations thereof, or various components thereof, may be examples of components used to perform aspects of the coding techniques for neural network architectures described herein. For example, the communication manager 1220, receiver 1210, transmitter 1215, or various combinations thereof, or components thereof, may support methods for performing one or more functions described herein.

[0183] In some examples, the communication manager 1220, receiver 1210, transmitter 1215, or various combinations or components thereof may be implemented in hardware (e.g., in communication management circuitry). The hardware may include a processor, DSP, ASIC, FPGA, or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof, configured or otherwise supporting components for performing the functions described herein. In some examples, the processor and memory coupled to the processor may be configured to perform one or more functions described herein (e.g., by the processor executing instructions stored in memory).

[0184] Additionally or alternatively, in some examples, the communication manager 1220, receiver 1210, transmitter 1215, or various combinations or components thereof may be implemented using processor-run code (e.g., as communication management software or firmware). If implemented as processor-run code, the functionality of the communication manager 1220, receiver 1210, transmitter 1215, or various combinations or components thereof may be performed by a general-purpose processor, DSP, CPU, ASIC, FPGA, or any combination of these or other programmable logic devices (e.g., components configured or otherwise supported for performing the functions described in this disclosure).

[0185] In some examples, the communication manager 1220 can be configured to use or cooperate with the receiver 1210, the transmitter 1215, or both, or otherwise, to perform various operations (e.g., receiving, monitoring, sending). For example, the communication manager 1220 can receive information from the receiver 1210, send information to the transmitter 1215, or integrate with the receiver 1210, the transmitter 1215, or a combination thereof to receive information, send information, or perform various other operations described herein.

[0186] According to the examples disclosed herein, the communication manager 1220 may support wireless communication at the device. For example, the communication manager 1220 may be configured or otherwise support components for transmitting to the UE an instruction for one or more encoding operations performed by the UE to encode a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both. The communication manager 1220 may be configured or otherwise support components for receiving from the UE an encoded, quantized, and compressed dataset after the compressed dataset has been encoded based on one or more encoding operations following a quantization operation. The communication manager 1220 may be configured or otherwise support components for decoding the encoded, quantized, and compressed dataset at least partially based on one or more encoding operations to generate a compressed dataset. The communication manager 1220 may be configured or otherwise support components for decoding the compressed dataset via a neural network to generate a dataset based on the decoded encoded, quantized, and compressed dataset.

[0187] Figure 13 A block diagram 1300 of a device 1305 supporting coding techniques for neural network architectures according to various aspects of this disclosure is shown. Device 1305 may be an example of device 1205 or aspects of base station 105 as described herein. Device 1305 may include receiver 1310, transmitter 1315, and communication manager 1320. Device 1305 may also include a processor. Each of these components can communicate with each other (e.g., via one or more buses).

[0188] Receiver 1310 may provide components 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 coding techniques used in neural network architectures). The information may be transmitted to other components of device 1305. Receiver 1310 may utilize a single antenna or a collection of antennas.

[0189] Transmitter 1315 may provide components for transmitting signals generated by other components of device 1305. For example, transmitter 1315 may transmit information associated with various information channels (e.g., control channels, data channels, information channels related to coding techniques used in neural network architectures), such as packets, user data, control information, or any combination thereof. In some examples, transmitter 1315 may be co-located with receiver 1310 in a transceiver module. Transmitter 1315 may utilize a single antenna or a collection of multiple antennas.

[0190] Device 1305 or its various components may be examples of parts for performing aspects of the encoding techniques for neural network architectures described herein. For example, communication manager 1320 may include encoding operation instruction component 1325, encoded dataset receiving component 1330, decoder component 1335, neural network decoder component 1340, or any combination thereof. Communication manager 1320 may be examples of aspects of communication manager 1220 as described herein. In some examples, communication manager 1320 or its various components may be configured to use receiver 1310, transmitter 1315, or both, or otherwise cooperate with them to perform various operations (e.g., receiving, monitoring, transmitting). For example, communication manager 1320 may receive information from receiver 1310, transmit information to transmitter 1315, or integrate with receiver 1310, transmitter 1315, or a combination thereof to receive information, transmit information, or perform various other operations described herein.

[0191] According to the examples disclosed herein, the communication manager 1320 may support wireless communication at the device. The encoding operation indication component 1325 may be configured or otherwise supported to provide a component for transmitting to the UE an indication of one or more encoding operations performed by the UE to encode a compressed dataset, including differential coding operations or entropy coding operations, or both. The encoded dataset receiving component 1330 may be configured or otherwise supported to provide a component for receiving from the UE an encoded, quantized, and compressed dataset that has already been encoded based on one or more encoding operations after a quantization operation. The decoder component 1335 may be configured or otherwise supported to provide a component for decoding the encoded, quantized, and compressed dataset based on one or more encoding operations to generate a compressed dataset. The neural network decoder component 1340 may be configured or otherwise supported to provide a component for decoding the compressed dataset via a neural network to generate a dataset based on the decoded encoded, quantized, and compressed dataset.

[0192] Figure 14 A block diagram 1400 of a communication manager 1420 supporting encoding techniques for neural network architectures according to various aspects of this disclosure is shown. As described herein, the communication manager 1420 may be an example of a communication manager 1220, a communication manager 1320, or aspects of both. The communication manager 1420 or its various components may be examples of parts for performing various aspects of the encoding techniques for neural network architectures described herein. For example, the communication manager 1420 may include an encoding operation instruction component 1425, an encoded dataset receiving component 1430, a decoder component 1435, a neural network decoder component 1440, a differential decoder component 1445, an entropy decoder component 1450, or any combination thereof. Each of these components may communicate with each other directly or indirectly (e.g., via one or more buses).

[0193] According to the examples disclosed herein, the communication manager 1420 may support wireless communication at the device. The encoding operation indication component 1425 may be configured or otherwise supported to provide a component for transmitting to the UE an indication of one or more encoding operations performed by the UE to encode a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both. The encoded dataset receiving component 1430 may be configured or otherwise supported to provide a component for receiving from the UE an encoded, quantized, and compressed dataset that has already been encoded based on one or more encoding operations after a quantization operation. The decoder component 1435 may be configured or otherwise supported to provide a component for decoding the encoded, quantized, and compressed dataset based on one or more encoding operations to generate a compressed dataset. The neural network decoder component 1440 may be configured or otherwise supported to provide a component for decoding the compressed dataset via a neural network to generate a dataset based on the decoded encoded, quantized, and compressed dataset.

[0194] In some examples, to support the sending of instructions for one or more encoded operations, the encoded operation instruction component 1425 may be configured or otherwise support components for sending one or more parameters corresponding to one or more encoded operations, each of the one or more parameters corresponding to a corresponding encoded operation of the one or more encoded operations, wherein the quantized and compressed dataset is encoded based on one or more parameters.

[0195] In some examples, to support decoding of encoded, quantized, and compressed datasets, the differential decoder component 1445 can be configured or otherwise supported as a component for decoding encoded, quantized, and compressed datasets using differential decoding operations.

[0196] In some examples, to support decoding of encoded, quantized, and compressed datasets, the entropy decoder component 1450 can be configured or otherwise supported as a component for decoding encoded, quantized, and compressed datasets using entropy decoding operations.

[0197] In some examples, to support receiving encoded, quantized, and compressed datasets, the differential decoder component 1445 may be configured or otherwise supported for receiving encoded, quantized, and compressed datasets that include differential values ​​of the data in the dataset based on the initial values ​​of the data.

[0198] In some examples, to support receiving encoded, quantized, and compressed datasets, the differential decoder component 1445 may be configured or otherwise supported for receiving encoded, quantized, and compressed datasets that include differential values ​​of previously reconstructed values ​​of data in the dataset.

[0199] In some examples, to support receiving encoded, quantized, and compressed datasets, the differential decoder component 1445 may be configured or otherwise supported for receiving encoded, quantized, and compressed datasets that include differential values ​​of the data based on the initial reconstructed values ​​of the data in the dataset.

[0200] In some examples, differential coding involves encoding a certain amount of data based on previous values ​​of a certain amount of data in a compressed dataset.

[0201] In some examples, entropy coding involves encoding a compressed dataset using one or more symbols of varying lengths based on the probability of their occurrence.

[0202] Figure 15 A diagram of a system 1500 including a device 1505 supporting coding techniques for neural network architectures is shown according to various aspects of this disclosure. Device 1505 may be an example of or include components of device 1205, device 1305, or base station 105 as described herein. Device 1505 may wirelessly communicate with one or more base stations 105, UE 115, or any combination thereof. Device 1505 may include components for bidirectional voice and data communication, including components for transmitting and receiving communications, such as a communication manager 1520, a network communication manager 1510, a transceiver 1515, an antenna 1525, a memory 1530, code 1535, a processor 1540, and an inter-station communication manager 1545. These components may communicate electronically or be otherwise coupled (e.g., operatively, communicatively, functionally, electronically, or electrically) via one or more buses (e.g., bus 1550).

[0203] The network communication manager 1510 can manage communication with the core network 130 (e.g., via one or more wired backhaul links). For example, the network communication manager 1510 can manage the transmission of data communication by client devices (e.g., one or more UEs 115).

[0204] In some cases, device 1505 may include a single antenna 1525. However, in other cases, device 1505 may have more than one antenna 1525, capable of transmitting or receiving multiple wireless transmissions simultaneously. Transceiver 1515 may communicate bidirectionally via one or more antennas 1525, wired or wireless links, as described herein. For example, transceiver 1515 may represent a wireless transceiver and may communicate bidirectionally with another wireless transceiver. Transceiver 1515 may also include a modem for modulating packets, providing modulated packets to one or more antennas 1525 for transmission, and demodulating packets received from one or more antennas 1525. As described herein, transceiver 1515 or transceiver 1515 and one or more antennas 1525 may be examples of transmitter 1215, transmitter 1315, receiver 1210, receiver 1310, or any combination thereof or components thereof.

[0205] Memory 1530 may include RAM and ROM. Memory 1530 may store computer-readable, computer-executable code 1535, including instructions that, when executed by processor 1540, cause device 1505 to perform the various functions described herein. Code 1535 may be stored in a non-transitory computer-readable medium, such as system memory or another type of memory. In some cases, code 1535 may not be directly executable by processor 1540, but may cause a computer (e.g., when compiled and run) to perform the functions described herein. In some cases, among others, memory 1530 may contain a BIOS that controls basic hardware or software operations, such as interaction with peripheral components or devices.

[0206] Processor 1540 may include intelligent hardware devices (e.g., general-purpose processors, DSPs, CPUs, microcontrollers, ASICs, FPGAs, programmable logic devices, discrete gate or transistor logic components, discrete hardware components, or any combination thereof). In some cases, processor 1540 may be configured to use a memory controller to operate a memory array. In other cases, the memory controller may be integrated into processor 1540. Processor 1540 may be configured to execute computer-readable instructions stored in memory (e.g., memory 1530) to cause device 1505 to perform various functions (e.g., functions or tasks supporting coding techniques for neural network architectures). For example, device 1505 or components thereof may include processor 1540 and memory 1530 coupled to processor 1540, processor 1540 and memory 1530 being configured to perform the various functions described herein.

[0207] Inter-site communication manager 1545 can manage communication with other base stations 105 and may include a controller or scheduler for cooperating with other base stations 105 to control communication with UE 115. For example, inter-site communication manager 1545 can coordinate the scheduling of transmissions to UE 115 for various interference suppression techniques such as beamforming or joint transmission. In some examples, inter-site communication manager 1545 may provide an X2 interface in LTE / LTE-A wireless communication network technology to facilitate communication between base stations 105.

[0208] According to the examples disclosed herein, the communication manager 1520 can support wireless communication at the device. For example, the communication manager 1520 can be configured or otherwise supported to support components for transmitting to the UE an instruction for one or more encoding operations performed by the UE to encode a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both. The communication manager 1520 can be configured or otherwise supported to support components for receiving from the UE an encoded, quantized, and compressed dataset after the compressed dataset has been encoded based on one or more encoding operations following a quantization operation. The communication manager 1520 can be configured or otherwise supported to support components for decoding the encoded, quantized, and compressed dataset at least partially based on one or more encoding operations to generate a compressed dataset. The communication manager 1520 can be configured or otherwise supported to support components for decoding the compressed dataset via a neural network to generate a dataset based on the decoded encoded, quantized, and compressed dataset.

[0209] In some examples, the communication manager 1520 may be configured to perform various operations (e.g., receiving, monitoring, transmitting) using or otherwise cooperating with a transceiver 1515, one or more antennas 1525, or any combination thereof. Although the communication manager 1520 is illustrated as a separate component, in some examples, one or more functions described with reference to the communication manager 1520 may be supported or performed by a processor 1540, memory 1530, code 1535, or any combination thereof. For example, code 1535 may include instructions that can be run by the processor 1540 to cause the device 1505 to perform aspects of the coding techniques for neural network architectures as described herein, or the processor 1540 and memory 1530 may be otherwise configured to perform or support such operations.

[0210] Figure 16 A flowchart of a method 1600 supporting encoding techniques for neural network architectures according to various aspects of this disclosure is shown. The operation of method 1600 can be implemented by a UE or its components as described herein. For example, the operation of method 1600 can be implemented by, as referenced... Figures 1 to 11The UE 115 described herein performs the following: In some examples, the UE may execute a set of instructions to control the functional elements of the UE to perform the described functions. Additionally or alternatively, the UE may use dedicated hardware to perform aspects of the described functions.

[0211] At 1605, the method may include receiving an instruction for encoding one or more encoding operations for encoding the compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both. The operation at 1605 may be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1605 may be performed by encoder instruction component 1025, as referenced... Figure 10 As described.

[0212] At 1610, the method may include encoding the dataset by a neural network to generate a compressed dataset. The operation at 1610 can be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1610 can be performed by the neural network encoder component 1030, as referenced... Figure 10 As described.

[0213] At 1615, the method may include quantizing a compressed dataset encoded by a neural network. The operation at 1615 can be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1615 can be performed by the quantization component 1035, as referenced... Figure 10 As described.

[0214] At 1620, the method may include encoding the quantized and compressed dataset based on instructions received for one or more encoding operations. The operations at 1620 can be performed according to the examples disclosed herein. In some examples, aspects of the operations at 1620 may be performed by encoder component 1040, as referenced... Figure 10 As described.

[0215] At 1625, the method may include transmitting the encoded, quantized, and compressed dataset to a second device after encoding the compressed dataset based on the one or more encoding operations. The operation at 1625 may be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1625 may be performed by the transmitting component 1045 of the encoded dataset, as referenced... Figure 10 As described.

[0216] Figure 17 A flowchart of a method 1700 supporting encoding techniques for neural network architectures according to various aspects of this disclosure is shown. The operation of method 1700 can be implemented by a UE or its components as described herein. For example, the operation of method 1700 can be implemented by, as referenced... Figures 1 to 11The UE 115 described herein performs the following: In some examples, the UE may execute a set of instructions to control the functional elements of the UE to perform the described functions. Additionally or alternatively, the UE may use dedicated hardware to perform aspects of the described functions.

[0217] At 1705, the method may include receiving an instruction for encoding one or more encoding operations for encoding the compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both. The operation at 1705 may be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1705 may be performed by encoder instruction component 1025, as referenced... Figure 10 As described.

[0218] At 1710, the method may include encoding the dataset by a neural network to generate a compressed dataset. The operation at 1710 can be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1710 can be performed by the neural network encoder component 1030, as referenced... Figure 10 As described.

[0219] At 1715, the method may include quantizing a compressed dataset encoded by a neural network. The operation at 1715 can be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1715 can be performed by the quantization component 1035, as referenced... Figure 10 As described.

[0220] At 1720, the method may include encoding the quantized and compressed dataset based on instructions received for one or more encoding operations. The operations at 1720 can be performed according to the examples disclosed herein. In some examples, aspects of the operations at 1720 may be performed by encoder component 1040, as referenced... Figure 10 As described.

[0221] At 1725, the method may include encoding the quantized and compressed dataset using differential coding operations. The operations at 1725 can be performed according to the examples disclosed herein. In some examples, aspects of the operations at 1725 can be performed by the differential encoder component 1050, as referenced... Figure 10 As described.

[0222] At 1730, the method may include transmitting the encoded, quantized, and compressed dataset to a second device after encoding the compressed dataset based on the one or more encoding operations. The operation at 1730 may be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1730 may be performed by the transmitting component 1045 of the encoded dataset, as referenced... Figure 10As described.

[0223] Figure 18 A flowchart of a method 1800 supporting encoding techniques for neural network architectures according to various aspects of this disclosure is shown. The operation of method 1800 can be implemented by a UE or its components as described herein. For example, the operation of method 1800 can be implemented by, as referenced... Figures 1 to 11 The UE 115 described herein performs the following: In some examples, the UE may execute a set of instructions to control the functional elements of the UE to perform the described functions. Additionally or alternatively, the UE may use dedicated hardware to perform aspects of the described functions.

[0224] At 1805, the method may include receiving an instruction for encoding one or more encoding operations for encoding the compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both. The operation at 1805 may be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1805 may be performed by encoder instruction component 1025, as referenced... Figure 10 As described.

[0225] At 1810, the method may include encoding the dataset by a neural network to generate a compressed dataset. The operation at 1810 can be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1810 can be performed by the neural network encoder component 1030, as referenced... Figure 10 As described.

[0226] At 1815, the method may include quantizing a compressed dataset encoded by a neural network. The operation at 1815 can be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1815 can be performed by the quantization component 1035, as referenced... Figure 10 As described.

[0227] At 1820, the method may include encoding the quantized and compressed dataset based on instructions received for one or more encoding operations. The operations at 1820 can be performed according to the examples disclosed herein. In some examples, aspects of the operations at 1820 may be performed by encoder component 1040, as referenced... Figure 10 As described.

[0228] At 1825, the method may include encoding the quantized and compressed dataset using an entropy coding operation. The operation at 1825 can be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1825 can be performed by the entropy encoder component 1055, as referenced... Figure 10 As described.

[0229] At 1830, the method may include transmitting the encoded, quantized, and compressed dataset to a second device after encoding the compressed dataset based on the one or more encoding operations. The operation at 1830 may be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1830 may be performed by the transmitting component 1045 of the encoded dataset, as referenced... Figure 10 As described.

[0230] Figure 19 A flowchart of a method 1900 supporting encoding techniques for neural network architectures according to various aspects of this disclosure is shown. The operation of method 1900 can be implemented by a base station or its components as described herein. For example, the operation of method 1900 can be implemented by [reference to...] Figures 1 to 7 and Figures 12 to 15 The described base station 105 performs the functions described. In some examples, the base station may run an instruction set to control the functional elements of the base station to perform the described functions. Additionally or alternatively, the base station may use dedicated hardware to perform aspects of the described functions.

[0231] At 1905, the method may include sending to the UE an instruction for one or more encoding operations for encoding a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both. The operation at 1905 may be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1905 may be performed by the encoding operation instruction component 1425, as referenced... Figure 14 As described.

[0232] At 1910, the method may include receiving from the UE a compressed dataset that has been encoded, quantized, and compressed after a quantization operation and based on one or more encoding operations. The operation at 1910 can be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1910 can be performed by the receiving component 1430 of the encoded dataset, as referenced... Figure 14 As described.

[0233] At 1915, the method may include decoding the encoded, quantized, and compressed dataset based on one or more encoding operations to generate a compressed dataset. The operation at 1915 can be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1915 may be performed by the decoder component 1435, as referenced... Figure 14 As described.

[0234] At 1920, the method may include generating a dataset by decoding a compressed dataset using a neural network based on decoding an encoded, quantized, and compressed dataset. The operation at 1920 can be performed according to the examples disclosed herein. In some examples, aspects of the operation at 1920 can be performed by the neural network decoder component 1440, as referenced... Figure 14 As described.

[0235] Figure 20 A flowchart of a method 2000 supporting encoding techniques for neural network architectures according to various aspects of this disclosure is shown. The operation of method 2000 can be implemented by a base station or its components as described herein. For example, the operation of method 2000 can be implemented by, as referenced... Figures 1 to 7 and Figures 12 to 15 The described base station 105 performs the functions described. In some examples, the base station may run an instruction set to control the functional elements of the base station to perform the described functions. Additionally or alternatively, the base station may use dedicated hardware to perform aspects of the described functions.

[0236] At point 2005, the method may include sending to the UE an instruction for one or more encoding operations by the UE to encode a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both. The operations at point 2005 may be performed according to the examples disclosed herein. In some examples, aspects of the operations at point 2005 may be performed by the encoding operation instruction component 1425, as referenced... Figure 14 As described.

[0237] At 2010, the method may include sending one or more parameters corresponding to one or more encoding operations, each of the one or more parameters corresponding to a corresponding encoding operation of the one or more encoding operations, wherein the quantized and compressed dataset is encoded based on the one or more parameters. The operation of 2010 may be performed according to the examples disclosed herein. In some examples, aspects of the operation of 2010 may be performed by the encoding operation instruction component 1425, as referenced... Figure 14 As described.

[0238] At point 2015, the method may include receiving from the UE a compressed dataset that has been encoded, quantized, and compressed after a quantization operation and based on one or more encoding operations. The operation at point 2015 can be performed according to the examples disclosed herein. In some examples, aspects of the operation at point 2015 may be performed by a receiving component 1430 of the encoded dataset, as referenced... Figure 14 As described.

[0239] At 2020, the method may include decoding an encoded, quantized, and compressed dataset based on one or more encoding operations to generate a compressed dataset. The operations at 2020 can be performed according to the examples disclosed herein. In some examples, aspects of the operations at 2020 may be performed by the decoder component 1435, as referenced... Figure 14 As described.

[0240] At point 2025, the method may include generating a dataset by decoding a compressed dataset using a neural network based on decoding an encoded, quantized, and compressed dataset. The operations at 2025 can be performed according to the examples disclosed herein. In some examples, aspects of the operations at 2025 can be performed by the neural network decoder component 1440, as referenced... Figure 14 As described.

[0241] The following provides an overview of the various aspects of this disclosure:

[0242] Aspect 1: A method for wireless communication at a UE, the method comprising: receiving an instruction for encoding one or more encoding operations for encoding a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both; encoding the dataset by a neural network to generate a compressed dataset; quantizing the compressed dataset encoded by the neural network; encoding the quantized and compressed dataset at least in part based on the received instruction for the one or more encoding operations; and transmitting the encoded, quantized and compressed dataset to a second device after encoding the compressed dataset at least in part based on the one or more encoding operations.

[0243] Aspect 2: According to the method of aspect 1, receiving an instruction to one or more encoding operations includes: receiving one or more parameters corresponding to one or more encoding operations, each of the one or more parameters corresponding to a corresponding encoding operation of the one or more encoding operations, wherein the quantized and compressed dataset is encoded at least in part based on the one or more parameters.

[0244] Aspect 3: The method according to any one of Aspects 1 to 2, wherein encoding the quantized and compressed dataset includes: encoding the quantized and compressed dataset using differential encoding operations after encoding the dataset using a neural network.

[0245] Aspect 4: The method according to any one of Aspects 1 to 3, wherein encoding the quantized and compressed dataset includes: encoding the quantized and compressed dataset using an entropy encoding operation after encoding the dataset using a neural network.

[0246] Aspect 5: The method according to any one of Aspects 1 to 4, wherein encoding the quantized and compressed dataset comprises: after quantizing the compressed dataset encoded by the neural network, determining a difference value between a first value of data in the quantized and compressed dataset at an initial time instance and a second value of data at a second time instance after the initial time instance, wherein the difference value is determined at least in part based on instructions of one or more encoding operations, and the quantized and compressed dataset is encoded at least in part based on the difference value.

[0247] Aspect 6: The method according to any one of Aspects 1 to 5, wherein encoding the quantized and compressed dataset comprises: after quantizing the compressed dataset encoded by the neural network, determining a difference value between a first reconstructed value of data in the quantized and compressed dataset at an initial time instance and a second reconstructed value of data at a second time instance after the initial time instance, wherein the difference value is determined at least in part based on instructions of one or more encoding operations, and the quantized and compressed dataset is encoded at least in part based on the difference value.

[0248] Aspect 7: The method according to any one of Aspects 1 to 6, wherein encoding the quantized and compressed dataset comprises: after quantizing the compressed dataset encoded by the neural network, determining an initial reconstruction value of the data in the quantized and compressed dataset at an initial time instance associated with encoding the dataset; and after quantization, determining a difference value between the additional reconstruction value of the data at an additional time instance following the initial time instance and the initial reconstruction value of the data, wherein the difference value is determined at least in part based on instructions of one or more encoding operations, and the quantized and compressed dataset is encoded at least in part based on the difference value.

[0249] Aspect 8: The method according to any one of Aspects 1 to 7, wherein the differential coding operation includes encoding a certain amount of data based at least in part on previous values ​​of a certain amount of data in the compressed dataset.

[0250] Aspect 9: The method according to any one of Aspects 1 to 8, wherein the entropy coding operation includes encoding a compressed dataset using one or more symbols having a length that varies at least in part based on the probability of the symbols occurring.

[0251] Aspect 10: A method for wireless communication at a device, the method comprising: sending to a UE an indication of one or more encoding operations for encoding a compressed dataset, the one or more encoding operations including differential coding or entropy coding or both; receiving from the UE an encoded, quantized and compressed dataset having been encoded at least in part based on one or more encoding operations after a quantization operation; decoding the encoded, quantized and compressed dataset at least in part based on one or more encoding operations to generate a compressed dataset; and decoding the compressed dataset by a neural network at least in part based on the decoding of the encoded, quantized and compressed dataset to generate a dataset.

[0252] Aspect 11: According to the method of aspect 10, the instruction to send one or more encoding operations includes: sending one or more parameters corresponding to one or more encoding operations, each of the one or more parameters corresponding to a corresponding encoding operation of the one or more encoding operations, wherein the quantized and compressed dataset is encoded at least in part based on the one or more parameters.

[0253] Aspect 12: The method according to any one of Aspects 10 to 11, wherein decoding the encoded, quantized and compressed dataset comprises: decoding the encoded, quantized and compressed dataset using a differential decoding operation before decoding the dataset using a neural network.

[0254] Aspect 13: The method according to any one of Aspects 10 to 12, wherein decoding the encoded, quantized and compressed dataset comprises: decoding the encoded, quantized and compressed dataset using an entropy decoding operation before decoding the dataset using the neural network.

[0255] Aspect 14: The method according to any one of Aspects 10 to 13, wherein receiving the encoded, quantized and compressed dataset comprises: receiving the encoded, quantized and compressed dataset, the encoded, quantized and compressed dataset comprising at least in part differential values ​​of the data in the dataset based on the initial values ​​of the data.

[0256] Aspect 15: The method according to any one of Aspects 10 to 14, wherein receiving the encoded, quantized and compressed dataset comprises: receiving the encoded, quantized and compressed dataset, the encoded, quantized and compressed dataset comprising at least in part differential values ​​of data in the dataset based on previously reconstructed values ​​of the data.

[0257] Aspect 16: The method according to any one of Aspects 10 to 15, wherein receiving the encoded, quantized and compressed dataset includes: receiving the encoded, quantized and compressed dataset, the encoded, quantized and compressed dataset including at least partially differential values ​​of the data in the dataset based on the initial reconstructed values ​​of the data.

[0258] Aspect 17: The method according to any one of Aspects 10 to 16, wherein the differential decoding operation includes encoding a certain amount of data based at least in part on previous values ​​of a certain amount of data in the compressed dataset.

[0259] Aspect 18: The method according to any one of Aspects 10 to 17, wherein the entropy decoding operation includes encoding a compressed dataset using one or more symbols having a length that varies at least in part based on the probability of the symbols occurring.

[0260] Aspect 19: An apparatus for wireless communication at a UE, the apparatus comprising a processor; a memory coupled to the processor; and instructions stored in the memory and operable by the processor to cause the apparatus to perform any of the methods of aspects 1 to 9.

[0261] Aspect 20: An apparatus for wireless communication at a UE, the apparatus comprising at least one component for performing the method of any one of aspects 1 to 9.

[0262] Aspect 21: A non-transitory computer-readable medium storing code for wireless communication at a UE, the code including instructions executable by a processor to perform any of the methods of aspects 1 to 9.

[0263] Aspect 22: An apparatus for wireless communication at a device, the apparatus comprising a processor; a memory coupled to the processor; and instructions stored in the memory and operable by the processor to cause the apparatus to perform the methods of any one of Aspects 10 to 18.

[0264] Aspect 23: An apparatus for wireless communication at a device, the apparatus comprising at least one component for performing the method of any one of Aspects 10 to 18.

[0265] Aspect 24: A non-transitory computer-readable medium storing code for wireless communication at a device, the code including instructions executable by a processor to perform the methods of any of Aspects 10 to 18.

[0266] It should be noted that the methods described herein describe possible implementations, and the operations and steps can be rearranged or modified, and other implementations are also possible. Furthermore, aspects of two or more methods can be combined.

[0267] While aspects of LTE, LTE-A, LTE-A Pro, or NR systems may be described for illustrative purposes, and the terms LTE, LTE-A, LTE-A Pro, or NR may be used in most of the description, the techniques described herein apply beyond LTE, LTE-A, LTE-A Pro, or NR networks. For example, the techniques described can be applied to a variety of other wireless communication systems, such as Ultra Mobile Broadband (UMB), IEEE 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20, Flash-OFDM, and other systems and radio technologies not explicitly mentioned herein.

[0268] The 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 referenced throughout this specification can be represented by voltage, current, electromagnetic waves, magnetic fields or particles, light fields or particles, or any combination thereof.

[0269] The various illustrative blocks and components described herein may be implemented or performed using a general-purpose processor, DSP, ASIC, CPU, FPGA or other programmable logic device, discrete gate or transistor logic, discrete hardware component, or any combination designed to perform the functions described herein. A general-purpose processor may be a microprocessor, but alternatively, it may be any processor, controller, microcontroller, or state machine. The processor may 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 combined with a DSP core, or any other such configuration).

[0270] The functions described herein can be implemented in hardware, software running on a processor, firmware, or any combination thereof. If implemented in software running on a processor, these functions can be stored or transmitted as one or more instructions or code on a computer-readable medium. Other examples and implementations are within the scope of this disclosure and the appended claims. For example, due to the nature of software, the functions described herein can be implemented using software running on a processor, hardware, firmware, hardwiring, or any combination thereof. Features implementing the functions can also be physically located in various locations, including being distributed such that portions of the functions are implemented in different physical locations.

[0271] Computer-readable media include non-transitory computer storage media and communication media, encompassing any medium that facilitates the transfer of a computer program from one place to another. A non-transitory storage medium can be any available medium accessible by a general-purpose or special-purpose computer. As an example, and without limitation, a non-transitory computer-readable medium may include RAM, ROM, electrically erasable programmable ROM (EEPROM), flash memory, disc-on-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 components in the form of instructions or data structures and is accessible by a general-purpose or special-purpose computer or a general-purpose or special-purpose processor. Furthermore, any connection is appropriately referred to as a computer-readable medium. For example, if software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of a computer-readable medium. The disks and optical discs used in this article include CDs, laser discs, optical discs, DVDs, floppy disks, and Blu-ray discs, where disks typically reproduce data magnetically, while optical discs reproduce data optically using lasers. Combinations of these are also included within the scope of computer-readable media.

[0272] As used herein, the word "or" in a list of items, including in the claims (e.g., a list of items beginning with phrases such as "at least one" or "one or more"), signifies a list of inclusions such that 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). Similarly, as used herein, the phrase "based on" should not be construed as referring to a closed set of conditions. For example, an example step described as "based on condition A" could be based on conditions A and B without departing from the scope of this disclosure. In other words, as used herein, the phrase "based on" should be interpreted in the same manner as the phrase "at least partially based on".

[0273] The term "determine" encompasses a wide variety of actions; therefore, "determine" can include calculation, operation, processing, derivation, investigation, lookup (e.g., by searching in a table, database, or other data structure), confirmation, etc. Furthermore, "determine" can include receiving (e.g., receiving information), accessing (e.g., accessing data in memory), etc. Additionally, "determine" can include resolving, selecting, establishing, and other similar actions.

[0274] In the accompanying drawings, similar components or features may have the same reference numerals. Furthermore, various components of the same type can be distinguished by a dash following the reference numeral and a second numeral to differentiate similar components. If only the first reference numeral is used in the specification, the description applies to any similar part having the same first reference numeral, regardless of the second or other subsequent reference numerals.

[0275] The description herein, illustrated with reference to the accompanying drawings, describes an example configuration and does not represent all examples that can be implemented or that are within the scope of the claims. The term "example" as used herein means "serving as an example, instance, or illustration," and not "preferred" or "superior to other examples." The detailed description includes specific details intended to provide an understanding of the described techniques. However, these techniques can be implemented without these specific details. In some cases, known structures and devices are shown in block diagram form to avoid obscuring the concept of the described examples.

[0276] The description provided herein is intended to enable those skilled in the art to make or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the scope of this disclosure. Therefore, this disclosure is not limited to the examples and designs described herein, but is consistent with the widest scope of the principles and novel features disclosed herein.

Claims

1. A method for performing wireless communication at a user equipment (UE), the method comprising: Receive an instruction for one or more encoding operations for encoding a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both; The compressed dataset is generated by encoding the dataset using a neural network. Quantize the compressed dataset encoded by the neural network; The quantized and compressed dataset is encoded after being encoded using the neural network, based at least in part on the instructions received from the one or more encoding operations. as well as After the compressed dataset is encoded at least in part based on the one or more encoding operations, the encoded, quantized, and compressed dataset is sent to the second device.

2. The method of claim 1, wherein receiving the indication of the one or more encoding operations comprises: Receive one or more parameters corresponding to the one or more encoding operations, each of the one or more parameters corresponding to a corresponding encoding operation of the one or more encoding operations, wherein the quantized and compressed dataset is encoded at least in part based on the one or more parameters.

3. The method of claim 1, wherein encoding the quantized and compressed dataset comprises: After encoding the dataset using the neural network, the quantized and compressed dataset is encoded using the differential encoding operation.

4. The method of claim 1, wherein encoding the quantized and compressed dataset comprises: After encoding the dataset using the neural network, the quantized and compressed dataset is encoded using the entropy encoding operation.

5. The method of claim 1, wherein encoding the quantized and compressed dataset comprises: After quantizing the compressed dataset encoded by the neural network, a difference value is determined between a first value of the data in the quantized and compressed dataset at an initial time instance and a second value of the data at a second time instance after the initial time instance, wherein the difference value is determined at least in part based on the indication of the one or more encoding operations, and the quantized and compressed dataset is encoded at least in part based on the difference value.

6. The method of claim 1, wherein encoding the quantized and compressed dataset comprises: After quantizing the compressed dataset encoded by the neural network, a difference value is determined between a first reconstructed value of the data in the quantized and compressed dataset at a first time instance and a second reconstructed value of the data at a second time instance after the first time instance, wherein the difference value is determined at least in part based on the indication of the one or more encoding operations, and the quantized and compressed dataset is encoded at least in part based on the difference value.

7. The method of claim 1, wherein encoding the quantized and compressed dataset comprises: After quantizing the compressed dataset encoded by the neural network, initial reconstructed values ​​of the data in the quantized and compressed dataset are determined at the initial time instance associated with encoding the dataset; as well as After the quantization, a difference value is determined between the additional reconstructed value of the data at an additional time instance following the initial time instance and the initial reconstructed value of the data, wherein the difference value is determined at least in part based on the indication of the one or more encoding operations, and the quantized and compressed dataset is encoded at least in part based on the difference value.

8. The method of claim 1, wherein the differential encoding operation comprises encoding the data at least in part based on previous values ​​of a portion of the data in the compressed dataset.

9. The method of claim 1, wherein the entropy encoding operation comprises encoding the compressed dataset using one or more symbols, the one or more symbols having a length that varies at least in part based on the probability of the symbol occurring.

10. A method for wireless communication at a device, the method comprising: Send to the user equipment (UE) an indication of one or more encoding operations for the UE to encode a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both; The compressed dataset received from the UE is an encoded, quantized, and compressed dataset that has been encoded at least in part based on the one or more encoding operations after the quantization operation; The encoded, quantized, and compressed dataset is decoded, at least in part, based on the one or more encoding operations, before being decoded using a neural network to generate a compressed dataset. as well as The neural network decodes the compressed dataset to generate a dataset, at least in part based on decoding the encoded, quantized, and compressed dataset.

11. The method of claim 10, wherein the instruction to send the one or more encoding operations comprises: Send one or more parameters corresponding to the one or more encoding operations, each of the one or more parameters corresponding to a corresponding encoding operation of the one or more encoding operations, wherein the quantized and compressed dataset is encoded at least in part based on the one or more parameters.

12. The method of claim 10, wherein decoding the encoded, quantized, and compressed dataset comprises: Before using the neural network to decode the dataset, differential decoding is used to decode the encoded, quantized, and compressed dataset.

13. The method of claim 10, wherein decoding the encoded, quantized, and compressed dataset comprises: Before using the neural network to decode the dataset, the encoded, quantized, and compressed dataset is decoded using an entropy decoding operation.

14. The method of claim 10, wherein receiving the encoded, quantized, and compressed dataset comprises: Receive the encoded, quantized, and compressed dataset, the encoded, quantized, and compressed dataset including at least in part the difference values ​​of the data in the dataset based on the initial values ​​of the data.

15. The method of claim 10, wherein receiving the encoded, quantized, and compressed dataset comprises: Receive the encoded, quantized, and compressed dataset, the encoded, quantized, and compressed dataset including at least in part the difference values ​​of the data in the dataset based on previously reconstructed values ​​of the data.

16. The method of claim 10, wherein receiving the encoded, quantized, and compressed dataset comprises: Receive the encoded, quantized, and compressed dataset, which includes at least in part differential values ​​of the data in the dataset based on the initial reconstructed values ​​of the data.

17. An apparatus for performing wireless communication at a user equipment (UE), the apparatus comprising: processor; Memory coupled to the processor; as well as Instructions stored in the memory and executable by the processor to cause the device to perform the following operations: Receive an instruction for one or more encoding operations for encoding a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both; The compressed dataset is generated by encoding the dataset using a neural network. Quantize the compressed dataset encoded by the neural network; The quantized and compressed dataset is encoded after being encoded using the neural network, based at least in part on the instructions received from the one or more encoding operations. as well as After the compressed dataset is encoded at least in part based on the one or more encoding operations, the encoded, quantized, and compressed dataset is sent to the second device.

18. The apparatus of claim 17, wherein the instruction for receiving the indication of the one or more encoded operations is executable by the processor to cause the apparatus to: Receive one or more parameters corresponding to the one or more encoding operations, each of the one or more parameters corresponding to a corresponding encoding operation of the one or more encoding operations, wherein the quantized and compressed dataset is encoded at least in part based on the one or more parameters.

19. The apparatus of claim 17, wherein the instructions for encoding the quantized and compressed dataset are executable by the processor such that the apparatus: After encoding the dataset using the neural network, the quantized and compressed dataset is encoded using the differential encoding operation.

20. The apparatus of claim 17, wherein the instructions for encoding the quantized and compressed dataset are executable by the processor such that the apparatus: After encoding the dataset using the neural network, the quantized and compressed dataset is encoded using the entropy encoding operation.

21. The apparatus of claim 17, wherein the instructions for encoding the quantized and compressed dataset are executable by the processor such that the apparatus: After quantizing the compressed dataset encoded by the neural network, a difference value is determined between a first value of the data in the quantized and compressed dataset at an initial time instance and a second value of the data at a second time instance after the initial time instance, wherein the difference value is determined at least in part based on the indication of the one or more encoding operations, and the quantized and compressed dataset is encoded at least in part based on the difference value.

22. The apparatus of claim 17, wherein the instructions for encoding the quantized and compressed dataset are executable by the processor such that the apparatus: After quantizing the compressed dataset encoded by the neural network, a difference value is determined between a first reconstructed value of the data in the quantized and compressed dataset at a first time instance and a second reconstructed value of the data at a second time instance after the first time instance, wherein the difference value is determined at least in part based on the indication of the one or more encoding operations, and the quantized and compressed dataset is encoded at least in part based on the difference value.

23. The apparatus of claim 17, wherein the instructions for encoding the quantized and compressed dataset are executable by the processor such that the apparatus: After quantizing the compressed dataset encoded by the neural network, initial reconstruction values ​​of the data in the quantized and compressed dataset are determined at the initial time instance associated with encoding the dataset; and After the quantization, a difference value is determined between the additional reconstructed value of the data at an additional time instance following the initial time instance and the initial reconstructed value of the data, wherein the difference value is determined at least in part based on the indication of the one or more encoding operations, and the quantized and compressed dataset is encoded at least in part based on the difference value.

24. An apparatus for wireless communication at a device, the apparatus comprising: processor; Memory coupled to the processor; as well as Instructions stored in the memory and executable by the processor to cause the device to perform the following operations: Send to the user equipment (UE) an indication of one or more encoding operations for the UE to encode a compressed dataset, the one or more encoding operations including differential coding operations or entropy coding operations or both; The compressed dataset received from the UE is an encoded, quantized, and compressed dataset that has been encoded at least in part based on the one or more encoding operations after the quantization operation; The encoded, quantized, and compressed dataset is decoded, at least in part, based on the one or more encoding operations, before being decoded using a neural network to generate a compressed dataset. as well as The neural network decodes the compressed dataset to generate a dataset, at least in part based on decoding the encoded, quantized, and compressed dataset.

25. The apparatus of claim 24, wherein the instruction for sending the indication of the one or more coded operations is executable by the processor to cause the apparatus to: Send one or more parameters corresponding to the one or more encoding operations, each of the one or more parameters corresponding to a corresponding encoding operation of the one or more encoding operations, wherein the quantized and compressed dataset is encoded at least in part based on the one or more parameters.

26. The apparatus of claim 24, wherein the instructions for decoding the encoded, quantized, and compressed dataset are executable by the processor, such that the apparatus: Before using the neural network to decode the dataset, differential decoding is used to decode the encoded, quantized, and compressed dataset.

27. The apparatus of claim 24, wherein the instructions for decoding the encoded, quantized, and compressed dataset are executable by the processor, such that the apparatus: Before using the neural network to decode the dataset, the encoded, quantized, and compressed dataset is decoded using an entropy decoding operation.

28. The apparatus of claim 24, wherein the instructions for receiving the encoded, quantized, and compressed dataset are executable by the processor to cause the apparatus to: Receive the encoded, quantized, and compressed dataset, the encoded, quantized, and compressed dataset including at least in part the difference values ​​of the data in the dataset based on the initial values ​​of the data.

29. The apparatus of claim 24, wherein the instructions for receiving the encoded, quantized, and compressed dataset are executable by the processor, such that the apparatus: Receive the encoded, quantized, and compressed dataset, the encoded, quantized, and compressed dataset including at least in part the difference values ​​of the data in the dataset based on previously reconstructed values ​​of the data.

30. The apparatus of claim 24, wherein the instructions for receiving the encoded, quantized, and compressed dataset are executable by the processor to cause the apparatus to: Receive the encoded, quantized, and compressed dataset, which includes at least in part differential values ​​of the data in the dataset based on the initial reconstructed values ​​of the data.

31. A computer-readable medium having program code recorded thereon, wherein the program code is executable by one or more processors to cause the one or more processors to perform the method of any one of claims 1-16.

Images