Hybrid wireless processing chain including deep neural network and static algorithm modules

Abstract

Techniques and apparatuses are described for a hybrid wireless communication processing chain including a deep neural network (DNN) and a static algorithm module. In an aspect, a first wireless communication device communicates with a second wireless device using a hybrid transmitter processing chain. The first wireless communication device selects a machine learning configuration (ML configuration) to form a modulated deep neural network (DNN) using coded bits as input to generate a modulated signal (805). The first wireless communication device forms a modulated DNN as part of a hybrid transmitter processing chain including the modulated DNN and at least one static algorithm module based on the modulated ML configuration (810). In response to forming the modulated DNN, the first wireless communication device processes wireless communications associated with the second wireless communication device using the hybrid transmitter processing chain (815).

Description

[Background technology]

[0001] background The evolution of wireless communication systems is often driven by demands for data throughput. As one example, the demand for data throughput increases as more and more devices gain access to wireless communication systems. As another example, evolving devices run data-intensive applications that utilize more data throughput than traditional applications, such as data-intensive streaming video applications, data-intensive social media applications, and data-intensive audio services. Such increased demand can sometimes exceed the available data throughput of a wireless communication system. Thus, to accommodate increased data usage, evolving wireless communication systems utilize increasingly complex architectures to provide more data throughput compared to legacy wireless communication systems. Summary of the Invention [Problem to be solved by the invention]

[0002] To increase data capacity, fifth-generation (5G) standards and technologies transmit data using higher frequency ranges, such as frequency bands above 6 gigahertz (GHz). However, transmitting and recovering information using these higher frequency ranges poses challenges. Higher frequency signals are more susceptible to multipath fading, scattering, atmospheric absorption, diffraction, interference, and the like than lower frequency radio signals. These signal distortions often lead to errors when recovering information at a receiver. User mobility also affects how well information can be transmitted and / or recovered using these higher frequency ranges, as channel conditions change as devices move from location to location. Hardware capable of transmitting, receiving, routing, and / or otherwise using these higher frequencies can be complex and expensive, which increases processing costs in wireless network devices. With recent technological advances, novel techniques may be available to improve the performance (e.g., data throughput, reliability) of wireless communications. [Means for solving the problem]

[0003] overview This document describes techniques and apparatus for a hybrid wireless communication processing chain including a deep neural network (DNN) and a static algorithm module. In an aspect, a first wireless communication device communicates with a second wireless device using a hybrid transmitter processing chain. The first wireless communication device selects a machine learning configuration (ML configuration) to form a modulated deep neural network (DNN) that uses coded bits as input to generate a modulated signal. The first wireless communication device forms the modulated DNN as part of a hybrid transmitter processing chain that includes the modulated DNN and at least one static algorithm module based on the modulated ML configuration. Using the hybrid transmitter processing chain, the first wireless communication device transmits a wireless communication signal to the second wireless communication device.

[0004] In an aspect, a first wireless communication device communicates with a second wireless communication device using a hybrid receiver processing chain. The first wireless communication device selects a demodulation machine learning (ML) configuration to form a demodulation deep neural network (DNN) that uses a modulated signal as an input and generates coded bits as an output. The first wireless communication device uses the demodulation ML configuration to form a demodulation DNN as part of a hybrid receiver processing chain that includes at least one static algorithm module and the demodulation DNN. Using the hybrid receiver processing chain, the first wireless communication device processes wireless signals received from the second wireless communication device.

[0005] In an aspect, a base station communicates with a user equipment (UE) using a hybrid wireless communication processing chain including at least one DNN and at least one static algorithm module. The base station selects a machine learning configuration (ML configuration) to form a base station-side DNN (e.g., a base station-side modulation DNN) that uses coded bits as input to generate a modulated downlink signal or uses a modulated uplink signal as input to generate coded bits. The base station indicates the ML configuration to the UE, and forms the base station-side DNN as part of the hybrid wireless communication processing chain including the base station-side DNN and at least one static algorithm based on the indicated ML configuration. The base station processes wireless communications using the hybrid wireless communication processing chain.

[0006] In an aspect, a UE communicates with a base station in a wireless network using a wireless communication processing chain including a DNN and at least one static algorithm module. The UE receives an indication of an ML configuration to form a DNN to process wireless communications associated with the base station. The UE then selects a UE-side ML configuration to form a UE-side DNN that (i) uses a modulated downlink signal as an input to generate coded bits as an output, or (ii) uses the coded bits as an input to generate a modulated uplink signal. The UE then uses the UE-side ML configuration to form a UE-side DNN as part of a hybrid wireless communication processing chain including the at least one static algorithm module and the UE-side DNN, and uses the hybrid wireless communication processing chain to process wireless communications associated with the base station.

[0007]

[0013] Details of one or more implementations of a hybrid wireless communication processing chain including a DNN and static algorithm module are set forth in the accompanying drawings and the following description. Other features and advantages will become apparent from the description and drawings, and from the claims. This Summary is provided to introduce the subject matter that is further described in the Detailed Description and Drawings. As such, this Summary should not be considered to describe essential features, nor should it be used to limit the scope of the claimed subject matter.

[0008] Details of one or more aspects of a hybrid wireless communication processing chain including a deep neural network (DNN) and static algorithm module are described below. The use of the same reference numbers in different instances within the description and drawings indicates similar elements. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 illustrates an example environment in which various aspects of a hybrid wireless communication processing chain including a DNN and static algorithm module may be implemented. [Figure 2]1 is an example device diagram of a device that may implement various aspects of a hybrid wireless communication processing chain including a DNN and static algorithm module. [Figure 3] FIG. 10 illustrates an example of generating multiple neural network formation configurations according to an embodiment of a hybrid wireless communication processing chain including a DNN and a static algorithm module. [Figure 4] FIG. 1 illustrates an example embodiment comparing downlink processing chains for wireless communications according to various aspects of a hybrid wireless communication processing chain including a DNN and a static algorithm module. [Figure 5] 1 is an example transaction diagram between various network entities implementing a hybrid wireless communication processing chain including DNN and static algorithm modules. [Figure 6] 1 is an example transaction diagram between various network entities implementing a hybrid wireless communication processing chain including DNN and static algorithm modules. [Figure 7] 1 is an example transaction diagram between various network entities implementing a hybrid wireless communication processing chain including DNN and static algorithm modules. [Figure 8] FIG. 1 illustrates a first example method for a hybrid wireless communication processing chain including a DNN and a static algorithm module. [Figure 9] FIG. 10 illustrates a second example method for a hybrid wireless communication processing chain including a DNN and a static algorithm module. [Figure 10] FIG. 10 illustrates a third example method for a hybrid wireless communication processing chain including a DNN and a static algorithm module. [Figure 11]FIG. 10 illustrates a fourth example method for a hybrid wireless communication processing chain including a DNN and a static algorithm module. DETAILED DESCRIPTION OF THE INVENTION

[0010] Detailed Description To accommodate increased data usage, evolving wireless communication systems (e.g., fifth-generation (5G) systems, sixth-generation (6G) systems) utilize higher frequency ranges and increasingly complex architectures to provide greater data throughput compared to legacy wireless communication systems. Illustratively, higher radio frequencies may add complexity to the transmitter and receiver processing chains to successfully exchange data wirelessly using the higher frequency ranges. For example, a channel estimation block in the receiver processing chain estimates or predicts how the transmission environment will distort a signal propagating through that transmission environment. A channel equalizer block inverts distortions identified from the signal by the channel estimation block. These complex functions often become more complex when processing higher frequency ranges, such as 5G frequencies in, around, and / or above the 6 GHz range. For example, the transmission environment adds more distortion to higher frequency ranges compared to lower frequency ranges, making information recovery more complex. User mobility introduces dynamic changes to the transmission environment as mobile devices move from location to location, which also contributes to the complexity of transmitting and recovering information using higher frequency ranges. For example, distortions introduced into a signal propagating toward a first location are different from distortions introduced into a signal propagating toward a second location. Hardware capable of processing and routing higher frequency ranges adds increased cost and complex physical constraints to the device.

[0011] Deep neural networks (DNNs) provide solutions to complex processing, such as complex functionality used within wireless communication systems. By training a DNN for wireless communication processing chain operations (e.g., transmitter and / or receiver processing chain operations), the DNN can replace the conventional complex functionality in various ways, such as by replacing some or all of the conventional processing blocks used in end-to-end processing of wireless communication signals, by replacing individual wireless communication processing chain blocks (e.g., modulation blocks, demodulation blocks), etc. Dynamic reconfiguration of DNNs, such as by modifying various machine learning configurations (e.g., coefficients, layer connections, kernel sizes), also provides the ability to adapt to changing operating conditions, such as user mobility, interference from neighboring cells, and bursty traffic.

[0012] The complexity of implementing and / or training a DNN increases with various factors, such as the complexity and / or amount of functionality provided by the DNN, the number of input parameters to the DNN, the variation and / or range of the input parameters, the amount and / or range of variation in the training data, etc. For example, a first DNN providing most or all of the functionality included in a wireless communications signal processing chain may be more complex than a second DNN providing a subset of the functionality included in the wireless communications signal processing chain. As examples, the first DNN may process larger amounts of training data, process larger amounts of input data, use more system computational power and / or memory, use longer durations for training and / or real-time computation, etc., compared to the second DNN.

[0013] Machine learning algorithms (e.g., DNNs) dynamically modify models or algorithms, while traditional algorithms use predefined rules. As one example, traditional encoders and / or decoders encode and / or decode bits using static and / or fixed algorithms. This may include static algorithms implemented using any combination of software, firmware, and / or hardware. Illustratively, traditional encoders (and / or decoders) implement static encoding algorithms (and / or static decoding algorithms) by explicitly programming predefined logic and / or rules to be used under all operating conditions. Similarly, static encoding algorithms produce the same output given the same input. The predefined logic and / or rules may use input parameters (e.g., encoding / decoding rates) to configure features and / or select specific program branches of the algorithm to vary the output. However, the input parameters do not modify or change the predefined logic and / or rules. In contrast, machine learning algorithms (e.g., DNNs) use training and feedback to dynamically modify the algorithm's behavior and / or resulting output. For example, machine learning algorithms identify patterns in data through training and feedback and generate new logic that modifies the machine learning algorithm to predict or identify these patterns in new (future) data.

[0014] In an embodiment of a hybrid wireless communications processing chain including a DNN and a static algorithm module, a device implements the hybrid wireless communications processing chain (e.g., a hybrid transmitter processing chain and / or a hybrid receiver processing chain) using a combination of DNNs and static algorithms to balance complexity and adaptability. The inclusion of a trained DNN in the wireless communications processing chain provides adaptability to changing input data and operating environments, such as dynamic changes in wireless communications due to user mobility, interference, multiple-input multiple-output (MIMO) configurations, etc. The inclusion of a static algorithm in the wireless communications chain reduces the amount of complexity in the trained DNN by reducing the amount of functionality provided by the DNN. In other words, using a combination of static algorithms and DNNs within the wireless communications processing chain reduces implementation complexity and provides adaptability to changing channel environments. As an example, the base station and / or UE may use static bit encoding and / or decoding algorithms within the wireless communications processing chain to reduce design and / or implementation complexity (e.g., by using conventional encoders / decoders), and may use modulation and / or demodulation DNNs (e.g., DNNs trained to perform modulation and demodulation) to increase the adaptability of the processing chain to dynamic operating environments (e.g., changing channel conditions, changing network loads, changing UE locations, changing UE data requirements). Alternatively or additionally, the modulation and / or demodulation DNNs may be trained to perform various MIMO operations, such as antenna selection, MIMO precoding, MIMO spatial multiplexing, MIMO diversity coding processing, MIMO spatial recovery, and MIMO diversity recovery. This combination helps simplify the complexity of the DNNs while maintaining the adaptability provided through the use of DNNs.

[0015] Example Environment 1 illustrates an example environment 100 including user equipment 110 (UE 110) that may communicate with base station 120 (illustrated as base stations 121 and 122) over one or more wireless communication links 130 (wireless link 130), illustrated as wireless links 131 and 132. For simplicity, UE 110 is implemented as a smartphone, but may be implemented as any suitable computing or electronic device, such as a mobile communication device, a modem, a mobile phone, a gaming device, a navigation device, a media device, a laptop computer, a desktop computer, a tablet computer, a smart appliance, a vehicle-based communication system, or an Internet of Things (IoT) device such as a sensor or actuator. The base station 120 (e.g., an Evolved Universal Terrestrial Radio Access Network Node B, E-UTRAN Node B, evolved Node B, eNodeB, eNB, Next Generation Node B, gNode B, gNB, ng-eNB, or the like) may be implemented in a macro cell, micro cell, small cell, pico cell, distributed base station, and the like, or any combination or future development thereof.

[0016] Base station 120 communicates with user equipment 110 using wireless links 131 and 132, which may be implemented as any suitable type of wireless link. Wireless links 131 and 132 include control and data communications, such as a downlink of data and control information communicated from base station 120 to user equipment 110, an uplink of other data and control information communicated from user equipment 110 to base station 120, or both. Wireless link 130 may include one or more wireless links (e.g., radio links) or bearers implemented using any suitable communications protocol or standard, or combination of communications protocols or standards, such as 3rd Generation Partnership Project Long-Term Evolution (3GPP® LTE), Fifth Generation New Radio (5G NR), and future developments. In various aspects, the base station 120 and the UE 110 are implemented for operation in sub-GHz bands, sub-6 GHz bands (e.g., Frequency Range 1), and / or frequency bands above 6 GHz (e.g., Frequency Range 2, millimeter wave (mmWave) bands) defined by one or more of the 3GPP LTE, 5G NR, or 6G communication standards (e.g., 26 GHz, 28 GHz, 38 GHz, 39 GHz, 41 GHz, 57-64 GHz, 71 GHz, 81 GHz, 92 GHz bands, 100 GHz-300 GHz, 130 GHz-175 GHz, or 300 GHz-3 THz bands). Multiple wireless links 130 may be aggregated into carrier aggregation or multiple connectivity to provide higher data rates for the UE 110. Multiple wireless links 130 from multiple base stations 120 may be configured for Coordinated Multipoint (CoMP) communication with the UE 110.

[0017] The base stations 120 are collectively a radio access network 140 (e.g., RAN, Evolved Universal Terrestrial Radio Access Network, E-UTRAN, 5G NR RAN, NR RAN). Base stations 121 and 122 within the RAN 140 are connected to a core network 150. The base stations 121 and 122 connect to the core network 150 at 102 and 104, respectively, through an NG2 interface for control plane signaling and using an NG3 interface for user plane data communication when connecting to a 5G core network, or using an S1 interface for control plane signaling and user plane data communication when connecting to an Evolved Packet Core (EPC) network. The base stations 121 and 122 may communicate at 106 using the Xn Application Protocol (XnAP) over an Xn interface or the X2 Application Protocol (X2AP) over an X2 interface to exchange user plane and control plane data. User equipment 110 may connect to a public network, such as the Internet 160 , via core network 150 to interact with remote services 170 .

[0018] Example Device 2 illustrates an example device diagram 200 of one of a UE 110 and a base station 120 that may implement various aspects of a hybrid wireless communication processing chain including DNN and static algorithm modules. The UE 110 and the base station 120 may include additional functionality and interfaces that are omitted from FIG. 2 for purposes of clarity.

[0019] The UE 110 includes an antenna array 202, a radio frequency front end 204 (RF front end 204), and one or more wireless transceivers 206 (e.g., an LTE transceiver, a 5G NR transceiver, and / or a 6G transceiver) for communicating with base stations 120 in the RAN 140. The RF front end 204 of the UE 110 may couple or connect the wireless transceiver 206 to the antenna array 202 to facilitate various types of wireless communications. The antenna array 202 of the UE 110 may include an array of multiple antennas configured similarly or differently from each other. The antenna array 202 and the RF front end 204 may be tuned and / or tunable to one or more frequency bands specified by the 3GPP LTE communication standard, the 5G NR communication standard, the 6G communication standard, and / or various satellite frequency bands, such as L-band (1-2 gigahertz (GHz)), S-band (2-4 GHz), C-band (4-8 GHz), X-band (8-12 GHz), Ku-band (12-18 GHz), K-band (18-27 GHz), and / or Ka-band (27-40 GHz), and implemented by the wireless transceiver 206. In some aspects, the satellite frequency bands overlap with the frequency bands of the 3GPP LTE, 5G NR, and / or 6G standards. Additionally, the antenna array 202, the RF front end 204, and / or the wireless transceiver 206 may be configured to support beamforming for transmitting and receiving communications with the base station 120. By way of example and not limitation, the antenna array 202 and RF front end 204 may be implemented for operation in sub-gigahertz (GHz), sub-6 GHz, and / or above 6 GHz frequency bands defined by 3GPP LTE, 5G NR, 6G, and / or satellite communications (e.g., satellite frequency bands).

[0020] The UE 110 also includes one or more processors 208 and a computer-readable storage medium 210 (CRM 210). The processor 208 may be a single-core processor or a multiple-core processor constructed of various materials, such as silicon, polysilicon, high-k dielectrics, copper, etc. The computer-readable storage media described herein exclude propagating signals. The CRM 210 may include any suitable memory or storage device, such as random access memory (RAM), static RAM (SRAM), dynamic RAM (DRAM), non-volatile RAM (NVRAM), read-only memory (ROM), or flash memory, that can be used to store device data 212 of the UE 110. The device data 212 may include user data, sensor data, control data, automation data, multimedia data, beamforming codebooks, applications, and / or the operating system of the UE 110, some of which are executable by the processor 208 to enable user plane data, control plane information, and user interaction with the UE 110.

[0021] In an embodiment, CRM 210 includes a neural network table 214 that stores various architecture and / or parameter configurations that form a neural network, such as, by way of example and not limitation, a fully connected neural network architecture, a convolutional neural network architecture, a recurrent neural network architecture, several connected hidden neural network layers, an input layer architecture, an output layer architecture, several nodes utilized by the neural network, coefficients (e.g., weights and biases) utilized by the neural network, kernel parameters, several filters utilized by the neural network, parameters specifying the stride / pooling configuration utilized by the neural network, activation functions for each neural network layer, interconnections between neural network layers, neural network layers to skip, etc. Thus, neural network table 214 includes any combination of neural network formation components (NN formation components), such as architectures and / or parameter configurations, that can be used to create a neural network formation configuration (NN formation configuration). Generally, a NN formation configuration includes a combination of one or more NN formation components that define and / or form a DNN. In some embodiments, a single index value in the neural network table 214 maps to a single NN formation component (e.g., a 1:1 correspondence). Alternatively, or additionally, a single index value in the neural network table 214 maps to a NN formation configuration (e.g., a combination of NN formation components). In some implementations, the neural network table includes input features for each NN formation component and / or NN formation configuration, which describe properties about the training data used to generate the NN formation component and / or NN formation configuration as further described. In embodiments, a machine learning configuration (ML configuration) corresponds to a NN formation configuration.

[0022] The CRM 210 may also include a user equipment neural network manager 216 (UE neural network manager 216). Alternatively, or additionally, the UE neural network manager 216 may be implemented, in whole or in part, as hardware logic or circuitry integrated with or separate from other components of the user equipment 110. The UE neural network manager 216 accesses the neural network table 214, such as as index values, and forms the DNN using NN formation components specified by the NN formation configuration, such as modulating and / or demodulating DNNs. This includes updating the DNN with any combination of architectural and / or parameter changes to the DNN as further described, such as small changes to the DNN involving updating parameters and / or large changes that reconfigure the node and / or layer connections of the DNN. In an implementation, the UE neural network manager forms multiple DNNs to process wireless communications, such as a first DNN forming a user equipment side demodulation deep neural network (UE side demodulation DNN) that receives analog-to-digital converter (ADC) samples of the (modulated) downlink signal as input and processes the ADC samples to recover the coded bits, and a second DNN forming a UE side modulation DNN that receives the coded bits as input and generates digital samples of the modulated baseband uplink signal or digital samples of a modulated intermediate frequency (IF) signal carrying the coded bits. In some aspects, the UE neural network manager 216 forwards updated machine learning parameters, such as those generated by a training module, to the base station 120 to contribute information for federated learning, as further described with reference to FIG. 8.

[0023] The CRM 210 includes a user equipment training module 218 (UE training module 218). Alternatively, or additionally, the UE training module 218 may be implemented, in whole or in part, as hardware logic or circuitry integrated with or separate from other components of the user equipment 110. The UE training module 218 teaches and / or trains the DNN using known input data and / or using feedback. As one example, the UE training module 218 trains the UE-side demodulation DNN using a cyclic redundancy check (CRC), as further described with reference to FIGS. 4 and 6. For illustrative purposes, assume that the UE-side demodulation DNN receives ADC samples of a downlink signal as input and processes the ADC samples to recover coded bits. The UE training module 218 may train the UE-side demodulation DNN by adjusting various ML parameters (e.g., weights, biases) based on CRC pass or fail. However, the UE training module 218 may alternatively or additionally train the UE-side demodulation DNN. The UE training module 218 may train the DNN offline (e.g., while the DNN is not actively involved in processing wireless communications) and / or online (e.g., while the DNN is actively involved in processing wireless communications).

[0024] The UE 110 also includes one or more static algorithm modules 220. The static algorithm modules 220 may be implemented using any combination of hardware, software, and / or firmware. Thus, the static algorithm modules 220 may be implemented using processor-executable instructions stored in the CRM 210 and executable by the processor 208 (not shown in FIG. 2). Generally, static algorithm modules perform various types of operations using pre-defined logic and / or rules that do not change. In an aspect, the static algorithm modules 220 implement operations associated with a wireless communication processing chain, such as encoding and / or decoding algorithms.

[0025] The device diagram for the base station 120 shown in FIG. 2 includes a single network node (e.g., gNode B). The functionality of the base station 120 may be distributed across multiple network nodes or devices, and may be distributed in any manner suitable for performing the functions described herein. The nomenclature of this distributed base station functionality varies and includes terms such as central unit (CU), distributed unit (DU), baseband unit (BBU), remote radio head (RRH), radio unit (RU), and / or remote radio unit (RRU). The base station 120 includes an antenna array 252, a radio frequency front end 254 (RF front end 254), and one or more wireless transceivers 256 (e.g., one or more LTE transceivers, one or more 5G NR transceivers, and / or one or more 6G transceivers) for communicating with the UEs 110. The RF front end 254 of the base station 120 may couple or connect the wireless transceiver 256 to the antenna array 252 to facilitate various types of wireless communications. The antenna array 252 of the base station 120 may include an array of multiple antennas configured similarly or differently. The antenna array 252 and the RF front end 254 may be tuned and / or tunable to one or more frequency bands defined by 3GPP LTE, 5G NR, 6G communication standards, and / or various satellite frequency bands and implemented by the wireless transceiver 256. Additionally, the antenna array 252, the RF front end 254, and the wireless transceiver 256 may be configured to support beamforming (e.g., massive multiple-input multiple-output (Massive-MIMO)) for transmitting and receiving communications to and from the UE 110.

[0026] The base station 120 also includes a processor 258 and a computer-readable storage medium 260 (CRM 260). The processor 258 may be a single-core processor or a multiple-core processor constructed of various materials, such as silicon, polysilicon, high-k dielectrics, copper, etc. The CRM 260 may include any suitable memory or storage device, such as random access memory (RAM), static RAM (SRAM), dynamic RAM (DRAM), non-volatile RAM (NVRAM), read-only memory (ROM), or flash memory, that can be used to store device data 262 of the base station 120. The device data 262 may include network scheduling data, radio resource management data, beamforming codebooks, applications, and / or the operating system of the base station 120, executable by the processor 258 to enable communication with the UE 110.

[0027] CRM 260 includes a neural network table 264 that stores a plurality of different NN formation components and / or NN formation configurations (e.g., ML configurations), where the NN formation components and / or NN formation configurations define various architectures and / or parameters for the DNN, as further described with reference to FIG. 5. In some implementations, the neural network table includes input features for each NN formation component and / or NN formation configuration, which describe properties about the training data used to generate the NN formation component and / or NN formation configuration. For example, input features may include, by way of example and not limitation, estimated UE location, multiple-input multiple-output (MIMO) antenna configuration, power information, signal-to-interference-plus-noise ratio (SINR) information, channel quality indicator (CQI) information, channel state information (CSI), Doppler feedback, frequency band, block error rate (BLER), quality of service (QoS), hybrid automatic repeat request (HARQ) information (e.g., first transmission error rate, second transmission error rate, maximum retransmission), latency, radio link control (RLC), automatic repeat request (ARQ) metrics, received signal strength (RSS), uplink SINR, timing management, error metrics, UE capability, BS capability, power mode, internet protocol (IP) layer throughput, end-to-end latency, end-to-end packet loss rate, etc. Thus, input features sometimes include layer 1, layer 2, and / or layer 3 metrics. In some implementations, a single index value in the neural network table 264 maps to a single NN forming component (e.g., a 1:1 correspondence). Alternatively, or additionally, a single index value in the neural network table 264 maps to a NN forming configuration (e.g., a combination of NN forming components).

[0028] In an implementation, base station 120 synchronizes neural network table 264 with neural network table 214 such that NN formation components and / or input features stored in one neural network table are replicated in the second neural network table. Alternatively, or additionally, base station 120 synchronizes neural network table 264 with neural network table 214 such that NN formation components and / or input features stored in one neural network table represent complementary functionality in the second neural network table. Illustratively, index values ​​that map to NN formation components forming a base station side modulation DNN (BS side modulation DNN) in neural network table 264 also map to NN formation components forming a (complementary) user equipment side demodulation DNN (UE side demodulation DNN) in neural network table 214.

[0029] CRM 260 also includes a base station neural network manager 266 (BS neural network manager 266). Alternatively, or additionally, BS neural network manager 266 may be implemented, in whole or in part, as hardware logic or circuitry integrated with or separate from other components of base station 120. In at least some aspects, BS neural network manager 266 selects the NN formation configuration utilized by base station 120 and / or UE 110 to configure deep neural networks for processing wireless communications, such as by selecting a combination of NN formation components to form a BS-side modulation DNN for processing downlink communications, a base station-side demodulation deep neural network (BS-side demodulation DNN) for processing uplink communications, a user equipment-side demodulation deep neural network (UE-side demodulation DNN) for processing downlink communications, and / or a user equipment-side modulation DNN (UE-side modulation DNN) for processing uplink communications. In some implementations, the BS neural network manager 266 receives feedback from the UE 110 (e.g., a UE-selected NN formation configuration and / or a UE-selected DNN configuration) and selects the NN formation configuration based on the feedback. Alternatively or additionally, the BS neural network manager 266 uses the feedback to train the BS-side DNN. In some aspects, the BS neural network manager 266 uses federated learning techniques, as described with reference to FIG. 8, to identify a common NN formation configuration and / or a common ML configuration for multiple UEs.

[0030] The CRM 260 includes a base station training module 268 (BS training module 268). Alternatively, or additionally, the BS training module 268 may be implemented, in whole or in part, as hardware logic or circuitry integrated with or separate from other components of the base station 120. In an aspect, the BS training module 268 teaches and / or trains the DNN using known input data and / or using feedback. As one example, the BS training module 268 trains the BS-side modulation DNN using hybrid automatic repeat request (HARQ) information and / or feedback from the UE 110. For illustrative purposes, it is assumed that the BS-side modulation DNN receives coded bits of a downlink signal as input and generates digital signal samples corresponding to a modulated baseband downlink signal. However, in other aspects, the BS-side modulation DNN generates a digitally modulated IF downlink signal. The BS training module 268 may train the BS-side modulation DNN by adjusting various ML parameters (e.g., weights, biases) based on the HARQ information feedback. However, the BS training module 268 may alternatively or additionally train a BS-side demodulation DNN for processing uplink signals. The BS training module 268 may train the DNN offline (e.g., while the DNN is not actively involved in processing wireless communications) and / or online (e.g., while the DNN is actively involved in processing wireless communications).

[0031] In an embodiment, the BS training module 268 extracts learned parameter configurations from the DNN, as further described with reference to Figure 3. The BS training module 268 may then use the extracted learned parameter configurations to create and / or update the neural network table 264. The extracted parameter configurations include any combination of information that specifies the behavior of the neural network, such as node connections, coefficients, activation layers, weights, biases, pooling, etc.

[0032] The CRM 260 also includes a base station manager 270. Alternatively, or additionally, the base station manager 270 may be implemented, in whole or in part, as hardware logic or circuitry integrated with or separate from other components of the base station 120. In at least some aspects, the base station manager 270 configures the wireless transceiver 256 for communication with the UE 110.

[0033] The base station 120 also includes one or more static algorithm modules 272. The static algorithm module 220 may be implemented using any combination of hardware, software, and / or firmware. Thus, the static algorithm module 272 may be implemented using processor-executable instructions stored in the CRM 260 and executable by the processor 258 (not shown in FIG. 2). Generally, a static algorithm module performs various types of operations using pre-defined logic and / or rules that do not change. In an aspect, the static algorithm module 272 implements operations associated with a wireless communications processing chain, such as encoding and / or decoding algorithms.

[0034] The base station 120 also includes a core network interface 274 that configures the base station manager 270 to exchange user plane data, control plane information, and / or other data / information with core network functions and / or entities. As one example, the base station 120 uses the core network interface 274 to communicate with the core network 150 of FIG. 1.

[0035] Training and building deep neural networks Generally, a DNN corresponds to a group of connected nodes organized into three or more layers, and the DNN uses training and feedback to dynamically modify the behavior and / or resulting output of the DNN algorithm. For example, a DNN identifies patterns in data through training and feedback and generates novel logic that modifies the machine learning algorithm (implemented as a DNN) to predict or identify these patterns in new (future) data. The connected nodes between layers can be configured in various ways, such as a partially connected configuration in which a first subset of nodes in a first layer are connected to a second subset of nodes in a second layer, or a fully connected configuration in which each node in the first layer is connected to each node in the second layer. The nodes can use various algorithms and / or analyses to generate output information based on adaptive learning, such as single linear regression, multiple linear regression, logistic regression, stepwise regression, binary classification, multi-class classification, multivariate adaptive regression splines, and locally estimated scatterplot smoothing (LOESS). Sometimes, the algorithms include weights and / or coefficients that change based on adaptive learning, so that the weights and / or coefficients reflect the information learned by the DNN.

[0036] DNNs may also employ various architectures that determine which nodes in the corresponding neural network are connected, how data is advanced and / or retained within the neural network, what weights and coefficients are used to process input data, how data is processed, and so on. Collectively, these various factors describe the NN configuration (also referred to as a machine learning (ML) configuration). By way of example, a recurrent neural network (RNN), such as a long-short-term memory (LSTM) neural network, forms cycles between node connections to retain information from previous portions of an input data sequence. The recurrent neural network then uses the retained information for subsequent portions of the input data sequence. As another example, a feedforward neural network passes information to forward connections without forming cycles to retain information. While described in the context of node connections, it should be understood that the NN configuration may include various parameter configurations that affect how the neural network processes input data.

[0037] The NN formation configuration used to form a DNN may be characterized by various architectures and / or parameter configurations. For illustrative purposes, consider an example in which the DNN implements a convolutional neural network. Generally, a convolutional neural network corresponds to a type of DNN in which layers process data using convolution operations to filter input data. Thus, by way of example and not limitation, the convolutional NN formation configuration may be characterized by pooling parameters (e.g., specifying a pooling layer to reduce the dimensionality of the input data), kernel parameters (e.g., a filter size and / or kernel type to use in processing the input data), weights (e.g., biases used to classify the input data), and / or layer parameters (e.g., layer connections and / or layer types). Although described in the context of pooling parameters, kernel parameters, weight parameters, and layer parameters, other parameter configurations may be used to form a DNN. Thus, the NN formation configuration (e.g., an ML configuration) may include any other type of parameter that can be applied to a DNN that affects how the DNN processes input data to generate output data.

[0038] 3 illustrates an example 300 that describes an aspect of generating multiple NN formation configurations by a hybrid wireless communication processing chain including DNN and static algorithm modules. At times, various aspects of example 300 are implemented by any combination of UE neural network manager 216, UE training module 218, BS neural network manager 266, and / or BS training module 268 of FIG. 2.

[0039] The upper portion of FIG. 3 includes a DNN 302, which represents any suitable DNN that can be used to implement a hybrid wireless communications processing chain including a DNN and a static algorithm module, such as a modulation DNN and / or a demodulation DNN. In an embodiment, a neural network manager generates different NN formation configurations and / or ML configurations for the DNNs implementing portions of the wireless communications processing chain. Alternatively, or additionally, the neural network manager generates the NN formation configurations and / or ML configurations based on different transmission environments, transmission channel conditions, and / or MIMO configurations. The training data 304 represents example inputs to the DNN 302, such as data corresponding to digitally modulated baseband signals for any combination of downlink communications, uplink communications, MIMO and / or operating configurations, and / or transmission environments. In other embodiments, the training data 304 represents coded bits as described with reference to FIGS. 4 and 5. In some implementations, the training module mathematically generates the training data or accesses a file that stores the training data. Otherwise, the training module acquires real-world communications data. Thus, the training module may use mathematically generated data, static data, and / or real-world data to train the DNN 302. Some implementations generate input features 306 that describe various qualities of the training data, such as operating configuration, transmission channel metrics, MIMO configuration, UE capabilities, UE location, modulation scheme, coding scheme, etc.

[0040] The DNN 302 analyzes the training data and generates an output 308, represented here as binary data. However, in other embodiments, such as when the training data corresponds to coded bits, the output 308 corresponds to a digitally modulated baseband or IF signal. Some implementations repeatedly train the DNN 302 using the same set of training data and / or additional training data having the same input features to improve the accuracy of the machine learning module. During training, the machine learning module modifies some or all of the architecture and / or parameter configurations of the neural network included in the machine learning module, such as node connections, coefficients, kernel sizes, etc. Some embodiments of training include supplemental inputs (not shown in FIG. 3 ), such as soft-decoded inputs for training the demodulation DNN.

[0041] In an embodiment, the training module extracts an architecture and / or parameter configuration 310 (e.g., pooling parameters, kernel parameters, layer parameters, weights) for the DNN 302, such as when the training module identifies that the accuracy meets or exceeds a desired threshold, that the training process meets or exceeds a number of iterations, etc. The extracted architecture and / or parameter configuration from the DNN 302 corresponds to a NN formation configuration, a NN formation component, an ML configuration, and / or an update to the ML configuration. The architecture and / or parameter configuration may include any combination of fixed architecture and / or parameter configurations and / or variable architecture and / or parameter configurations.

[0042] The bottom of Figure 3 includes a neural network table 312 that represents a collection of NN formation components, such as neural network table 214 and / or neural network table 264 of Figure 2. The neural network table 312 stores various combinations of architecture configurations, parameter configurations, and input features, although alternative implementations omit input features from the table. Various implementations update and / or maintain the NN formation components and / or input features as the DNN learns additional information. For example, index 314, a neural network manager, and / or a training module update the neural network table 312 to include the architecture and / or parameter configurations 310 generated by the DNN 302 while analyzing the training data 304. At a later point in time, the neural network manager (e.g., UE neural network manager 216, BS neural network manager 266) selects one or more NN formation configurations from the neural network table 312 by matching the input features to the current operating environment and / or configuration, e.g., by matching the input features to the current channel conditions and / or MIMO configuration (e.g., antenna selection). In an aspect, the base station 120 communicates an indicator 314 to the UE 110 (or vice versa) to indicate which NN formation configuration should be used to form (e.g., generate, instantiate, or load) the DNN, as further described.

[0043] Hybrid wireless communication processing chain including DNN and static algorithm modules In embodiments of a hybrid wireless communication processing chain including a DNN and a static algorithm module, a device implements a wireless communication processing chain (e.g., a transmitter processing chain and / or a receiver processing chain) using a combination of a DNN and a static algorithm to balance complexity and adaptability. Each processing chain includes, for example, an encoding module and / or a decoding module that uses a static algorithm to perform modulation and / or demodulation operations and at least one DNN. The inclusion of a DNN provides flexibility to modify how transmissions are generated in response to changes in the operating environment, such as modulation scheme changes, channel condition changes, and MIMO configuration changes. Illustratively, some embodiments dynamically modify the DNN to generate transmissions with properties (e.g., frequency, modulation scheme, beam direction, MIMO antenna selection) that mitigate issues in the current transmission channel. For example, the inclusion of static algorithms through the static encoding module and / or static decoding module simplifies the complexity of the DNN (e.g., reduces processing time and reduces training time), balancing complexity and efficiency.

[0044] 4 illustrates a first example environment 400 and a second example environment 402 comparing wireless communication processing chains, where the processing chains include one or more DNNs, sometimes in combination with static algorithms, according to various aspects of a hybrid wireless communication processing chain including DNN and static algorithm modules. Environment 400 and environment 402 each include an example transmitter processing chain and an example receiver processing chain that may be used to process downlink (DL) wireless communications (e.g., a DL transmitter processing chain at base station 120, a DL receiver processing chain at UE 110) or uplink (UL) wireless communications (e.g., a UL transmitter processing chain at UE 110, a UL receiver processing chain at base station 120).

[0045] In environment 400, a BS neural network manager 266 (not shown in FIG. 4 ) of base station 120 manages one or more deep neural networks 404 (DNNs 404) included in a base station downlink processing chain 406 (BS downlink processing chain 406). In an aspect, BS neural network manager 266 configures DNNs 404 to perform transmitter processing chain operations for downlink wireless communications directed to UEs 110. Illustratively, BS neural network manager 266 selects one or more default ML configurations or one or more specific ML configurations (e.g., based on current downlink channel conditions, as further described) and configures DNNs 404 using the ML configurations. In an aspect, the DNN 404 performs some or all functionality of a (wireless communications) transmitter processing chain, such as receiving binary data as input, encoding the binary data, generating a digitally modulated baseband or IF signal using the encoded data, performing MIMO transmission operations (e.g., antenna selection, MIMO precoding, MIMO spatial multiplexing, MIMO diversity coding processing), and / or generating an upconverted signal (e.g., a digital representation) that feeds a digital-to-analog converter (DAC) that feeds the antenna array 252 for downlink transmission 408. Illustratively, the DNN 404 may perform any combination of convolutional coding, serial-to-parallel conversion, cyclic prefix insertion, channel coding, time / frequency interleaving, orthogonal frequency division multiplexing (OFDM), MIMO transmission operations, etc.

[0046] A UE neural network manager 216 (not shown in FIG. 4 ) of the UE 110 manages one or more deep neural networks 410 (DNNs 410) included in a user equipment downlink processing chain 412 (UE downlink processing chain 412). In an aspect, the UE neural network manager 216 configures the DNNs 410 to process downlink wireless communication signals received from the base station 120. Illustratively, the UE neural network manager 216 forms the DNNs 410 using an ML configuration indicated by the base station 120 and / or using an NN formation configuration selected by the UE neural network manager 216. In an aspect, the DNNs 410 perform some or all of the functionality of the (wireless communication) receiver processing chain, such as processing that is complementary to the processing performed by the BS DL processing chain (e.g., downconversion stage, demodulation stage, decoding stage), regardless of whether the BS DL processing chain includes one or more DNNs, static algorithm modules, or both. To illustrate, the DNN 410 may perform any combination of demodulating / extracting data embedded in the received (RX) signal, recovering control information, recovering binary data, correcting data errors based on forward error correction applied in the transmitter block, extracting payload data from frames and / or slots, etc.

[0047] Similarly, the UE 110 includes a first user equipment uplink processing chain 414 (UE uplink processing chain 414) that processes uplink communications using one or more deep neural networks 416 (DNNs 416) configured and / or formed by the UE neural network manager 216. Illustratively, and as previously described with reference to the DNN 404, the DNN 416 performs any combination of (uplink) transmitter chain processing operations to generate an uplink transmission 418 directed to the base station 120.

[0048] The base station 120 includes a first base station uplink processing chain 420 (BS uplink processing chain 420) that processes (received) uplink communications using one or more deep neural networks 422 (DNNs 422) managed by the BS neural network manager 266. The DNNs 422 perform complementary processing (e.g., receiver chain processing operations as described with reference to the DNNs 410) performed by the UE UL processing chain, regardless of whether the UE UL processing chain includes one or more DNNs, static algorithm modules, or both.

[0049] In contrast, environment 402 illustrates an exemplary hybrid wireless communication processing chain that uses a combination of static algorithm modules and DNNs to process uplink and / or downlink wireless communications. For example, environment 402 includes a hybrid transmitter processing chain 424 that uses a combination of static algorithm modules and DNNs. For example, base station 120 uses hybrid transmitter processing chain 424 in place of BS DL DNN processing chain 406 or a conventional static algorithm BS-side DL processing chain, and / or UE 110 uses hybrid transmitter processing chain 424 in place of UE UL DNN processing chain 412 or a conventional static algorithm UE-side uplink processing chain. Environment 402 also includes a hybrid receiver processing chain 426 that uses a combination of static algorithm modules and DNNs within the wireless communication receiver processing chain. To illustrate, the UE 110 uses the hybrid receiver processing chain 426 in place of the UE DL DNN processing chain 412 or a conventional static algorithm UE-side downlink processing chain, and / or the base station 120 uses the hybrid receiver processing chain 426 in place of the BS UL DNN processing chain 420 or a conventional static algorithm BS-side UL processing chain.

[0050] The hybrid transmitter processing chain 424 includes an encoding module 428 implemented using a static algorithm that receives source bits 430 (e.g., from a protocol stack, not shown in FIG. 4 ) and generates coded bits using one or more static coding algorithms, such as a low-density parity check (LPDC) coding algorithm, a polar coding algorithm, a turbo coding algorithm, and / or a Viterbi coding algorithm. The hybrid transmitter processing chain 424 utilizes any combination of hardware, software, and / or firmware to implement the encoding module 428. In an aspect, the encoding module 428 receives input parameters (e.g., channel coding scheme parameters, rate matching parameters) that instruct the encoding module how to code the source bits 430. By using a static algorithm within the encoding module 428, the hybrid transmitter processing chain 424 can use an encoding module that is optimized for better performance (e.g., optimized for processing speed, optimized for physical and / or memory size).

[0051] The hybrid transmitter processing chain 424 also includes a modulation module 432 including one or more modulation DNNs 434 that modulate the coded bits received from the encoding module 428. Illustratively, the DNNs 434 correspond to a base station side deep neural network (BS side DNN) that modulates downlink communications, also referred to as a BS side modulation DNN, and / or a user equipment side deep neural network (UE side DNN) that modulates uplink communications, also referred to as a UE side modulation DNN. In some aspects, the BS neural network manager 266 of the base station 120 selects one or more modulation ML configurations to form the modulation DNNs 434. As one example, the BS neural network manager 266 selects a base station side modulation ML configuration (BS side modulation ML configuration) that processes downlink communications, such as that described with reference to FIG. 5. Alternatively or additionally, the BS neural network manager 266 selects a user equipment side modulation ML configuration (UE side modulation ML configuration) to transmit to the UE 110, such as that described with reference to FIG. 6. In some embodiments, the BS neural network manager 266 selects updates to the modulation ML configuration, such as by using federated learning techniques as described with reference to FIG.

[0052] The BS neural network manager 266 selects a modulation ML configuration (e.g., a BS-side modulation ML configuration, a UE-side modulation ML configuration) using any combination of factors. Illustratively, the BS neural network manager 266 selects a modulation configuration using factors such as, for example, the current operating state, the UE capabilities of the UE 110, the MIMO configuration (e.g., antenna selection), the modulation scheme, and the channel conditions. Illustratively, with respect to the MIMO configuration, the BS neural network manager may select a modulation ML configuration based on the MIMO transmit and receive antenna configuration, such as a 2×2 MIMO configuration corresponding to two transmit antennas and two receive antennas, or a 4×4 MIMO configuration corresponding to four transmit antennas and four receive antennas. As another example, the BS neural network manager may select a modulation ML configuration based on the modulation scheme.

[0053] In an aspect, and as described with reference to FIG. 5, a base station (e.g., base station 120) may indicate to UE 110 the modulation ML configuration selected by the BS neural network manager, such as by indicating the BS-side modulation ML configuration through a field in downlink control information (DCI) transmitted in a physical downlink control channel (PDCCH) message. As one example, the DCI may include a first field specifying a channel coding scheme and a second field specifying the modulation ML configuration. Alternatively or additionally, the second field specifies changes and / or updates to the modulation ML configuration, such as changes identified through federated learning techniques as described with reference to FIG. 7. However, in some aspects, base station 120 implicitly indicates the channel coding scheme instead of using the first field in the DCI. As another example, the base station indicates the modulation ML configuration by transmitting particular reference and / or pilot signals, such as a channel state information reference signal (CSI-RS), a demodulation reference signal (DMRS), and / or a phase tracking reference signal (PTRS) that are mapped to a particular modulation ML configuration. In some aspects, the base station 120 selects a modulated ML configuration from a fixed number of modulated ML configurations and / or a predetermined set of modulated ML configurations, such as a subset of ML configurations stored in a neural network table and / or codebook, and transmits a codebook index to the UE.

[0054] Alternatively, using complementary operations, the base station 120 may select a user equipment side demodulation machine learning configuration (UE side demodulation ML configuration) for processing downlink communications based on the BS side modulation ML configuration and indicate the UE side demodulation ML configuration to the UE in the DCI. As noted above, the instruction may represent an index to the neural network and / or a codebook set of ML configurations. For illustrative purposes, it is assumed that the base station 120 selects the UE side modulation ML configuration using any combination of UE-specific capabilities, UE-specific signal quality measurements, UE-specific link quality measurements, UE-specific MIMO configurations, etc. When using the same transmission channel for downlink and uplink communications, such as through time division duplex (TDD) transmission, the base station 120 may select the UE side demodulation ML configuration to form a (downlink) UE side demodulation DNN based on the (downlink) BS side modulation ML configuration, such as that described with reference to FIG. 5. Alternatively or additionally, base station 120 may implicitly or explicitly indicate the UE-side demodulation ML configuration for downlink processing when indicating the BS-side modulation ML configuration in the DCI.

[0055] The DNN 434 performs modulation and / or MIMO operations within the hybrid transmitter processing chain 424. For example, the modulation DNN 434 receives coded bits from the coding module 428 and generates a digitally modulated baseband signal (e.g., digital samples of the modulated baseband signal). However, alternative implementations generate digitally modulated IF signals that are processed in a manner similar to that described with respect to the baseband signal. The digitally modulated baseband signal may include MIMO communications that simultaneously transmit several signals through multiple antennas. For example, the modulation DNN 434 may generate modulated baseband signals for 2×2 MIMO communications, 4×4 MIMO communications, etc. Thus, in an aspect, the modulation DNN 434 generates modulated baseband signals that split and / or replicate coded data onto multiple data streams. Alternatively or additionally, the modulation DNN 434 performs other MIMO operations in generating the digitally modulated baseband signal, such as MIMO precoding, MIMO spatial multiplexing, and / or MIMO diversity coding.

[0056] In generating the digitally modulated baseband signal, the modulation DNN 434 applies a modulation scheme to the encoded data, such as an orthogonal frequency division multiplexing (OFDM) modulation format. The selected modulation ML configuration configures the modulation DNN 434 to perform processing to apply OFDM modulation to the encoded data, such as binary phase shift keying (BPSK) with OFDM, quadrature phase shift keying (QPSK) with OFDM, or 16-quadrature amplitude modulation (16-QAM) with OFDM. In some aspects, such as when the modulation DNN 434 corresponds to a BS-side modulation DNN for processing downlink transmissions, the base station (as the BS neural network manager 266) updates the modulation DNN 434 based on current operation and / or channel conditions. For example, and with reference to FIG. 5, the base station 120 trains the modulation DNN 434 using feedback from the UE. As another example, such as when the modulation DNN 434 corresponds to a UE-side modulation DNN for processing uplink transmissions, the UE trains the modulation DNN 434 as described with reference to FIG. 6. This allows the base station 120 and / or the UE to improve transmission as operating and / or channel conditions change.

[0057] Within the hybrid transmitter processing chain 424, a modulation DNN, which may correspond to a downlink BS-side modulation DNN or an uplink UE-side modulation DNN, generates a digitally modulated baseband signal (or a digital IF signal). The modulation module 432 feeds the digitally modulated baseband signal to a transmit radio frequency processing module 436 (TX RF processing module 436), which is connected to an antenna (e.g., antenna array 252 when operating within the base station 120 or antenna array 202 when operating within the UE 110). The TX RF processing module 436 includes any combination of hardware, firmware, and / or software used to output transmissions via an antenna. For example, the TX RF processing module 436 includes a DAC that receives the digitally modulated baseband signal from the modulation module 432 and generates an analog modulated baseband signal. The TX RF processing module 436 may alternatively or additionally include a signal mixer that upconverts the analog modulated baseband signal to a desired carrier frequency, which is then transmitted from an antenna (e.g., antenna array 252 for downlink transmissions or antenna array 202 for uplink transmissions).

[0058] In environment 402, hybrid receiver processing chain 426 uses a combination of DNN and static algorithm modules (implemented either using traditional static algorithms, using DNN, or using a hybrid approach) to perform complementary processing to transmitter processing chain 406. For example, base station 120 uses hybrid receiver processing chain 426 in place of BS UL processing chain 420 (e.g., BS-side UL processing chain) and / or UE 110 uses hybrid receiver processing chain 426 in place of UE DL processing chain 412 (e.g., UE-side downlink processing chain).

[0059] As one example, the UE 110 receives downlink communications and / or transmissions from the base station 120 using the antenna array 202, where the downlink communications may include MIMO communications. The antenna routes the (analog) received downlink transmissions to a receive radio frequency processing module 438 (RX RF processing module 438) included in the hybrid receiver processing chain 426. The RX RF processing module 438 converts the received analog signals to digitally modulated baseband signals. However, in an alternative implementation, the RX RF processing module 438 generates a digitally modulated IF signal that is processed in a manner similar to the digitally modulated baseband signal. For example, the RX RF processing module 438 includes a mixer that downconverts the downlink transmissions to an analog baseband signal and an ADC that digitizes the downconverted analog signal to generate the digitally modulated baseband signal. The RX RF processing module 438 then inputs the digitally modulated baseband signal to a demodulation module 440 (e.g., a UE-side demodulation module for processing downlink communications, a BS-side demodulation module for processing uplink communications) that includes one or more demodulation DNNs 442. Illustratively, the UE 110 forms the demodulation DNN 442 using the UE neural network manager 216, or the base station 120 forms the demodulation DNN 442 using the BS neural network manager 266. In an aspect, the demodulation DNN 442 performs complementary processing to the modulation DNN 434 included in the modulation module 432, such as receiving the digitally modulated baseband signal and processing it to recover the encoded data. This may include MIMO operations, such as MIMO spatial and / or diversity recovery, channel estimation, a channel equalizer function, etc.

[0060] The demodulation module 440 inputs the recovered encoded data to a decoding module 444, which uses a static algorithm to generate recovered bits 446. This may include the decoding module 444 receiving input parameters (e.g., channel coding scheme parameters, rate matching parameters), which instruct the decoding module on how to decode and generate the recovered bits 446. In some aspects, the decoding module 444 generates soft decoding information 448, such as log-likelihood ratio information, and inputs the soft decoding information to the demodulation DNN 440. Similar to the encoding module 428, the decoding module 444 may implement any combination of static decoding algorithms (e.g., LPDC decoding algorithm, polar decoding algorithm, turbo decoding algorithm, and / or Viterbi decoding algorithm). In some aspects, the hybrid receiver processing chain 426 uses feedback from the decoding module 444, such as CRC information, to trigger training of the demodulation DNN 442 and / or uses the feedback to train the demodulation DNN 442, such as that described with reference to FIG. 5.

[0061] The hybrid wireless communications processing chain (e.g., hybrid transmitter processing chain 424, hybrid receiver processing chain 426) provides a device capable of balancing complexity and adaptability in implementing and operating the processing chain. Including static algorithms in the wireless communications processing chain reduces the amount of functionality provided by the corresponding DNN, which reduces system computational power and / or memory consumption by the DNN and reduces the processing and / or training duration of the DNN. The inclusion of a DNN within the wireless communications processing chain (connected to the static algorithm module) provides adaptability to changing operating and / or channel conditions to mitigate channel and / or operating issues.

[0062] Signaling and Data Transaction Diagram 5-7 illustrate example signaling and data transaction diagrams between a base station and user equipment in accordance with one or more aspects of a hybrid wireless communication processing chain including a DNN and static algorithm module. The signaling and data transaction operations may be performed by the base station 120 and / or the UE 110 of FIG. 1 using aspects such as those described with reference to any of FIGS. 1-4. Hybrid transmitter and receiver processing chains are assumed for both the BS and the UE, although in some circumstances the transmitter may use a hybrid processing chain while the receiver uses a conventional DNN or hybrid processing chain. Similarly, the receiver may use a hybrid processing chain when the transmitter uses a conventional DNN or hybrid processing chain.

[0063] A first example of signaling and data transactions for a hybrid wireless communication processing chain including a DNN and a static algorithm module is illustrated by signaling and data transaction diagram 500 of Figure 5. In diagram 500, a base station (e.g., base station 120) and a UE (e.g., UE 110) exchange downlink wireless communications using a processing chain including a combination of a static algorithm module and a DNN, in accordance with one or more aspects of a hybrid wireless communication processing chain including a DNN and a static algorithm module.

[0064] As illustrated, at 505, the base station 120 selects a base station side modulation machine learning configuration (BS side modulation ML configuration) to form a base station side DNN (BS side DNN) included in a BS side transmitter processing chain using a combination of at least one DNN and at least one static algorithm module. Illustratively, the base station 120 selects a modulation ML configuration for a BS side modulation DNN (e.g., DNN 434) in a downlink transmitter processing chain (e.g., hybrid transmitter processing chain 424). The base station 120 selects the BS side modulation ML configuration using any combination of information. As one example, the base station 120 selects a default BS side modulation ML configuration to form a DL modulation DNN for generating broadcast transmissions. In other words, the base station 120 selects a modulation ML configuration to form a modulation DNN that generates modulated transmissions with characteristics aimed at general and / or unknown channel conditions, such as transmissions having a DL modulation scheme that is more robust across a range of different channel conditions compared to other modulation schemes. Alternatively or additionally, the base station 120 selects the BS-side modulation ML configuration from a fixed number of ML configurations and / or a predetermined set of ML configurations (e.g., a set of ML configurations known to both the base station 120 and the UE 110). In some aspects, the base station 120 selects the BS-side modulation ML configuration based on information specific to the UE 110, such as UE location information, signal quality measurements, link quality measurements, UE capabilities, etc.

[0065] At 510, assuming the UE has indicated UE capabilities including demodulation DNN formation, base station 120 indicates a selected BS-side modulation ML configuration to UE 110. Base station 120 indicates the BS-side modulation ML configuration, for example, using a first field in DCI for the PDCCH, and instructs UE 110 (implicitly or explicitly) to select a reciprocal UE-side demodulation ML as further described at 515. By indicating the selected BS-side ML configuration, base station 120 provides different UEs (e.g., different manufacturers, different UE capabilities) with information about how the corresponding BS-side DNN operates, thus enabling each UE to select its own complementary UE-side ML configuration, such as that described at 515. Alternatively, base station 120 indicates the UE-side demodulation ML configuration to UE 110 and instructs UE 110 (implicitly or explicitly) to form a demodulation DNN (using the UE-side demodulation ML configuration) for processing downlink communications.

[0066] In an aspect, the base station 120 indicates the channel coding scheme using a second field in the DCI (e.g., a new DCI format) in addition to the ML configuration (e.g., the BS-side modulation configuration, the UE-side demodulation configuration). As an example of indicating the ML configuration (e.g., the BS-side modulation ML configuration, the UE-side demodulation ML configuration), assume that the base station 120 and the UE 110 use common and / or synchronous mapping for a predetermined set of ML configurations. In an aspect, the base station 120 indicates a specific ML configuration from the predetermined set of ML configurations to the UE 110, such as by indicating an index value that maps to the specific ML configuration. Based on the common and / or synchronous mapping, the UE 110 identifies the indicated ML configuration from the predetermined set of ML configurations using the index value. In some aspects, the base station 120 indicates the BS-side modulation ML configuration by transmitting a specific pilot and / or reference signal. For example, the base station 120 transmits a specific CSI-RS, DMRS, and / or PTRS that map to a specific BS-side modulation ML configuration and / or index value.

[0067] At 515, the UE 110 selects a UE-side demodulation ML configuration for a UE-side DNN included in a UE-side receiver processing chain, the UE-side receiver processing chain using a combination of at least one DNN and at least one static algorithm module. Illustratively, the UE 110 identifies an indicated BS-side modulation ML configuration communicated by the base station at 510 by analyzing the PDCCH DCI and / or by identifying a received reference and / or pilot signal. Based on identifying the indicated BS-side modulation ML configuration, the UE selects a complementary ML configuration to form a demodulation DNN (e.g., demodulation DNN 442) within the hybrid receiver processing chain (e.g., hybrid receiver processing chain 426). In some aspects, such as when the indication corresponds to an index value, the UE 110 uses the index value to obtain the UE-side demodulation ML configuration from a codebook and / or a predetermined set of ML configurations.

[0068] Alternatively or additionally, UE 110 selects a UE-side demodulation ML configuration by analyzing performance metrics (e.g., bit error rate (BER), block error rate (BLER)) of multiple demodulation ML configurations. As one example, UE 110 forms an initial UE-side demodulation DNN using an initial ML configuration complementary to an indicated BS-side demodulation ML configuration. UE 110 acquires performance metrics of the UE-side demodulation DNN and determines that the performance metrics indicate degraded performance (e.g., the performance metrics associated with the initial ML configuration fail to meet a performance threshold). In response, UE 110 selects a second UE-side demodulation ML configuration based on the performance metrics. In other words, UE 110 selects a second UE-side demodulation ML configuration that forms a second UE-side demodulation DNN having a better performance metric (e.g., a performance metric that meets a performance threshold) than the initial UE-side demodulation DNN. Illustratively, UE 110 analyzes a set of demodulation ML configurations and selects a demodulation ML configuration from the set that is associated with the best performance metric. In some aspects, as part of analyzing and selecting a UE-side demodulation ML configuration, the UE 110 selects a matching channel demodulation scheme from those indicated by the base station (e.g., at 510). Alternatively or additionally, the UE 110 selects a UE-side demodulation ML configuration that forms a demodulation DNN that demodulates a particular modulation configuration using OFDM (e.g., BPSK with OFDM, QPSK with OFDM, 16-QAM), etc. Thus, the UE 110 may also select a UE-side demodulation ML configuration based on any combination of other factors, such as transport block size, frequency grant size, spatial grant size, time grant size, etc. In some aspects, the UE 110 selects a UE-side demodulation ML configuration based on UE capabilities.

[0069] Thus, at 520, the UE optionally (as indicated by the dashed line) indicates its UE-selected UE-side demodulation ML configuration to the base station 120. The UE 110 may indicate its UE-selected UE-side demodulation ML configuration using any suitable mechanism, such as by transmitting a specific sounding reference signal (SRS) that is mapped to the selected demodulation ML configuration and / or by including an indication of the selected demodulation ML configuration within a channel state information (CSI) communication.

[0070] At 525, base station 120 forms a BS-side modulation DNN, which may include base station 120 using the BS-side modulation ML configuration selected at 505, or optionally selecting a second BS-side modulation ML configuration (and / or updating the BS-side modulation DNN) based on receiving an indication of a UE-selected UE-side demodulation ML configuration. Similarly, at 530, UE 110 forms a UE-side demodulation DNN, which may include UE 110 using the UE-side demodulation ML configuration based on the indication from base station 120 at 510, or using the UE-selected UE-side demodulation ML configuration determined by UE 110 at 515.

[0071] At 535, the base station 120 processes one or more downlink communications using the encoding module and the BS-side modulation DNN formed at 525. In an aspect, the BS-side modulation DNN receives as input coded bits from a coding module implemented using a static algorithm (e.g., the encoding module 428) and outputs a digitally modulated baseband signal. This may additionally include the BS-side modulation DNN performing MIMO operations as further described with reference to FIG. 4. At 540, the base station 120 transmits the downlink communications to the UE 110, such as by converting the digitally modulated baseband signals to analog RF signals using a TX RF processing module (e.g., the TX RF processing module 436) coupled to the antenna.

[0072] At 545, the UE 110 processes the downlink communication using the decoding module and the UE-side demodulation DNN formed at 530. Illustratively, the UE 110 uses an RX RF processing module (e.g., RX RF processing module 438) to downconvert the transmitted downlink communication at 540 and generate a digitally modulated baseband signal as described with reference to FIG. 4. The UE-side demodulation DNN receives the digitally modulated baseband signal as an input and recovers the coded bits. In an aspect, the UE 110 uses a decoding module (e.g., decoding module 444) implemented with a static algorithm to generate the recovered bits. In some aspects, the UE-side demodulation DNN receives soft-decoded information (e.g., soft-decoded information 448) as an input from the decoding module.

[0073] At 550, the UE 110 optionally (indicated by a dashed line) transmits feedback to the base station 120. For example, the UE 110 transmits HARQ information to the base station 120 at 550, which may or may not trigger the base station 120 to train the BS-side modulation DNN to adjust weights, biases, etc.

[0074] Accordingly, at 555, the base station 120 optionally (indicated by a dashed line) trains the BS-side modulated DNN. Illustratively, the base station 120 determines to perform training when the HARQ information indicates a failure at the UE 110. Alternatively or additionally, the base station 120 uses the signal quality and / or link quality measurements returned by the UE 110 at 550 to trigger training of the BS-side modulated DNN, such as by comparing the signal quality and / or link quality measurements to a threshold indicative of an acceptable performance level and triggering training when the signal quality and / or link quality measurements do not meet the acceptable performance level. In some aspects, the base station uses the HARQ information to train the BS-side modulated DNN, such as by using the same set of coded bits and adjusting ML parameters and / or ML architecture until the HARQ information indicates an acceptable performance level. In response to training the BS-side modulated DNN, the base station 120 updates the BS-side modulated DNN as shown at 560 and processes subsequent downlink communications using the updated BS-side modulated DNN.

[0075] At 565, UE 110 optionally (indicated by a dashed line) trains the UE-side demodulation DNN. Illustratively, UE 110 uses the CRC information from the decoding module to determine when to trigger a training procedure, such as by monitoring the CRC information and triggering training when the CRC indicates a failure “N” times in a row, where “N” is a predetermined value. As one example, UE 110 uses ADC samples of the modulated baseband signal and the CRC information generated by the decoding module as feedback to adjust various ML parameters (e.g., weights, biases) and / or ML architecture until the CRC information indicates an acceptable performance level. For example, the UE 110, via a UE neural network manager (e.g., UE neural network manager 216) and / or a training module (e.g., UE training module 218), selects adjustments that reduce bit errors and / or improve bit recovery by adjusting the ML parameters of the UE-side demodulation DNN with the gradient value, using CRC pass / fail information, and / or measuring a cost function of CRC errors (e.g., minimizing CRC errors). In some aspects, the UE neural network manager and / or training module decides to train the UE-side demodulation DNN by using a cost function to determine when the performance of the UE-side demodulation DNN has degraded below a performance threshold.

[0076] In response to training the UE-side demodulation DNN, the UE 110 optionally (indicated by a dashed line) updates the UE-side demodulation DNN as shown at 570 and processes subsequent downlink communications using the updated UE-side demodulation DNN. Alternatively or additionally, the UE 110 optionally extracts updates to the UE-side demodulation DNN as described with reference to FIG. 3 and transmits an ML configuration update (e.g., a UE-side ML configuration update) to the base station 120 at 575. Alternatively or additionally, the base station 120 optionally (indicated by a dashed line) extracts updates to the BS-side demodulation DNN as shown in FIG. Draw and transmits the ML update to the UE 110 at 580 .

[0077] A second example of signaling and data transactions for a hybrid wireless communication processing chain including a DNN and a static algorithm module is illustrated by signaling and data transaction diagram 600 of Figure 6. In diagram 600, a base station (e.g., base station 120) and a UE (e.g., UE 110) exchange uplink wireless communications using a processing chain including a combination of a static algorithm module and a DNN, in accordance with one or more aspects of a hybrid wireless communication processing chain including a DNN and a static algorithm module.

[0078] At 605, the base station 120 selects a UE-side modulation ML configuration for the UE-side DNN. Illustratively, and as similarly described at 505 in FIG. 5, the base station 120 selects a modulation ML configuration for the UE-side modulation DNN (e.g., modulation DNN 434) that generates a digitally modulated baseband signal using encoded UL data as input. The base station selects a default ML configuration that forms a modulation DNN suitable for multiple different types of UEs, channel conditions, etc. Alternatively or additionally, the base station 120 selects the UE-side modulation ML configuration from a pre-defined set of ML configurations (e.g., a set of ML configurations known to both the base station 120 and the UE 110) and / or based on information specific to the UE 110, such as UE location information, signal quality measurements, link quality measurements, UE capabilities, etc.

[0079] At 610, the base station 120 transmits an indication of the UE-side modulation ML configuration to the UE 110. In one example, the base station 120 transmits the indication of the UE-side modulation ML configuration in a field of the DCI for the PUSCH. In an aspect, the base station 120 transmits as the indication an index value that maps to an entry in a codebook and / or points to a particular ML configuration from among a pre-defined set of ML configurations synchronized between the base station 120 and the UE 110. In some aspects, the base station 120 implicitly indicates the UE-side modulation ML configuration based on a reciprocity as described with reference to FIG. 4.

[0080] At 615, the UE 110 selects a UE-side modulation ML configuration. Illustratively, and as similarly described at 515 in FIG. 5, the UE 110 may identify an indicated UE-side modulation ML configuration communicated by the base station at 610 and form a UE-side modulation DNN (e.g., modulation DNN 434) using the indicated ML configuration. Alternatively or additionally, the UE 110 analyzes performance metrics of one or more downlink reference signals (e.g., DMRS, PTRS, CSI-RS) to select a UE-side modulation ML configuration from a predetermined set and / or codebook of ML configurations. Accordingly, at 620, the UE 110 optionally (as indicated by a dashed line) indicates the UE-selected UE-side modulation ML configuration to the base station 120.

[0081] At 625, the UE 110 forms a UE-side modulation DNN using either the indicated UE-side modulation ML configuration or a UE-selected UE-side modulation ML configuration. Similarly, at 630, the base station 120 forms a BS-side demodulation DNN that performs complementary processing to the UE-side modulation DNN, where the BS-side demodulation ML configuration may be based on the UE-side modulation ML configuration indicated at 610 or the UE-selected UE-side modulation ML configuration indicated at 620.

[0082] At 635, the UE 110 processes one or more uplink communications using a coding module and a UE-side modulation DNN. Illustratively, and as described with reference to FIG. 5 , the UE-side modulation DNN (e.g., modulation DNN 434) receives coded bits from a coding module that uses one or more static coding algorithms (e.g., coding module 514). The UE-side modulation DNN processes the coded bits and generates a digitally modulated baseband signal, where the processing may include performing MIMO operations. The UE-side modulation DNN inputs the digitally modulated baseband signal to a TX RF processing module that generates an upconverted analog modulated signal, and at 640, the UE 110 transmits the uplink communications using the upconverted analog modulated signal and one or more antennas (e.g., antenna array 202) of the UE.

[0083] At 645, the base station 120 processes the uplink communication using a decoding module and a BS-side demodulation DNN. In an aspect, the base station 120 includes a BS-side demodulation DNN in a receiver processing chain that includes a combination of a DNN and a static algorithm module, such as the BS uplink processing chain 524 of FIG. 5. Illustratively, an RX RF processing module in the receiver processing chain converts a received analog signal to a digitally modulated baseband signal. The BS-side demodulation DNN processes the digitally modulated baseband signal to recover the encoded data and inputs the recovered encoded data to a static decoding module to generate recovered bits.

[0084] At 650, the base station 120 optionally (shown in dashed lines) transmits feedback to the UE 110. As one example, the base station 120 transmits BER information, BLER information, and / or CRC information to the UE 110.

[0085] At 655, UE 110 optionally (indicated by a dashed line) trains the UE-side modulation DNN. In an aspect, UE 110 triggers and / or initiates a training procedure for the UE-side modulation DNN based on the feedback transmitted at 650. For example, UE 110 analyzes the BER and / or BLER and triggers training when the BER and / or BLER exceed an acceptable threshold level of error and / or exceed the acceptable threshold level of error “M” times, where “M” corresponds to any number. In some aspects, UE 110 triggers and / or initiates the training procedure based on a signal quality measurement and / or a link quality measurement, such as a signal quality and / or link quality measurement indicating an interference level exceeds another threshold. In response to training the UE-side modulation DNN, UE 110 optionally updates the UE-side modulation DNN as shown at 660 and processes subsequent uplink communications using the updated UE-side modulation DNN. Alternatively or additionally, the UE 110 optionally extracts an update to the UE-side modulation DNN as described with reference to FIG. 3 and transmits an ML configuration update to the base station 120 at 665.

[0086] At 670, the base station 120 optionally trains the BS-side demodulation DNN. Illustratively, and as similarly described at 560 in FIG. 5 , the base station 120 triggers training of the BS-side demodulation DNN by monitoring CRC information and triggering training when the CRC indicates a failure “N” times in a row, where “N” is an arbitrary value. In an aspect, the base station 120 trains the BS-side demodulation DNN using ADC samples of the modulated baseband signal and CRC information generated by the decoding module as feedback to adjust various ML parameters (e.g., weights, biases). In response to training the BS-side demodulation DNN, the base station 120 updates the BS-side demodulation DNN as shown at 675 and uses the updated BS-side demodulation DNN to process subsequent uplink communications.

[0087] A third example of signaling and data transactions for a hybrid wireless communication processing chain including a DNN and static algorithm module is illustrated by signaling and data transaction diagram 700 of Figure 7. In diagram 700, a base station (e.g., base station 120) uses federated learning techniques to manage DNN configurations of modulation and / or demodulation DNNs used within a processing chain including a combination of DNN and static algorithm modules in accordance with one or more aspects of a hybrid wireless communication processing chain including a DNN and static algorithm module. Aspects of diagram 700 may be implemented by a base station (e.g., base station 120) and at least two UEs (e.g., at least two UEs 110).

[0088] Generally, federated learning corresponds to a distributed training mechanism for machine learning algorithms. Illustratively, an ML manager (e.g., BS neural network manager 266) selects a baseline ML configuration and instructs multiple devices to create and train an ML algorithm using the baseline ML configuration. The ML manager then receives and aggregates training results from the multiple devices to generate an updated ML configuration for the ML algorithm. As one example, multiple devices each report learned parameters (e.g., weights or coefficients) generated by the ML algorithm while processing their own specific input data, and the ML manager generates the updated ML configuration by averaging the weights or coefficients to create the updated ML configuration. As another example, multiple devices each report gradient results based on their own individual input data to the ML manager indicating an optimal ML configuration based on function processing cost (e.g., processing time, processing accuracy), and the ML manager averages the gradients. In some embodiments, multiple devices report learned ML architecture updates and / or changes from the baseline ML configuration. The terms federated learning, distributed training, and / or distributed learning may be used interchangeably.

[0089] At 705, the base station 120 selects a group of UEs. As one example, the base station 120 selects a group of UEs based on common UE capabilities, such as a common number of antennas or common transmit / receive power. Alternatively, or additionally, the base station 120 selects a group of UEs based on comparable signal or link quality measurements (e.g., parameters having values ​​within a threshold relative to each other). TeU E Group of This may include comparable uplink and / or downlink signal quality measurements, such as Reference Signal Received Power (RSRP), Signal-to-Interference-Plus-Noise Ratio (SINR), Channel Quality Indicator (CQI), etc. Based on any combination of common UE capabilities, comparable signal or link quality measurements, estimated UE locations (e.g., within a predetermined distance of each other), etc., base station 120 selects two or more UEs to include in a group for federated learning.

[0090] At 710, the base station 120 selects an initial ML configuration for a DNN included in a processing chain that utilizes a combination of DNN and static algorithm modules. Illustratively, the base station 120 selects the initial ML configuration for any combination of BS-side modulation DNN, UE-side demodulation DNN, UE-side modulation DNN, and / or BS-side demodulation DNN, as described with reference to Figures 4-6. Thus, the base station 120 may select multiple initial ML configurations, each corresponding to a different DNN.

[0091] At 715, base station 120 indicates an initial ML configuration to each of the UEs included in the group of UEs selected at 705. In other words, base station 120 indicates a common ML configuration to each of the UEs as the initial ML configuration. This may include indicating the initial ML configuration using DCI, CSI-RS, pilot signals, etc., as described with reference to FIGS. 4-7. In some aspects, base station 120 indicates to each of the UEs that the initial ML configuration corresponds to a baseline ML configuration for federated learning.

[0092] At 720, the base station 120 optionally (indicated by a dashed line) indicates one or more training conditions to each of the UEs included in the group of UEs selected at 705, the training conditions corresponding to triggering training of the corresponding DNN. Illustratively, the base station requests the UEs to report updated ML information (and / or perform a training procedure) by indicating one or more update conditions specifying rules or instructions regarding when to report updated ML information. As one example of an update condition, the base station 120 requests each UE in the group of UEs to periodically transmit updated ML information (and / or perform a training procedure) and indicates a recurrence time interval. As another example of an update condition, the base station 120 requests each UE in the group of UEs to transmit updated ML information (and / or perform a training procedure) in response to detecting a trigger event, such as a trigger event corresponding to a change in the DNN in the UE. To illustrate, base station 120 requests each UE to transmit updated ML information when the UE determines that an ML parameter (e.g., a weight or coefficient) has changed beyond a threshold. As another example, base station 120 requests each UE to transmit updated ML information in response to detecting when the ML architecture changes at the UE, such as when the UE (as UE neural network manager 216) identifies that the DNN has modified the ML architecture by adding or removing a node or layer.

[0093] In some aspects, the base station 120 requests the UE to report updated ML information based on signal or link quality measurements observed by the UE. Illustratively, the base station 120 requests the UE to report updated ML information in response to identifying, as a triggering event and / or update condition, that a downlink signal and / or link quality parameter (e.g., RSSI, SINR, CQI, channel delay spread, Doppler spread) has changed by or meets a threshold. As another example, the base station 120 requests the UE to report updated ML information in response to detecting a positive / negative acknowledgment (ACK / NACK) threshold. Thus, the base station 120 can request synchronous updates (e.g., periodic) from a group of UEs or asynchronous updates from a group of UEs based on conditions detected at each UE. In an aspect, the base station requests the UE to report observed signal or link quality measurements along with updated ML information.

[0094] At 725, the base station 120 and the UEs 110 included in the group process communications using respective processing chains including at least one DNN and at least one static algorithm module. Illustratively, and referring to FIG. 4, the base station 120 processes downlink communications using a hybrid transmitter processing chain (e.g., hybrid transmitter processing chain 424) including a BS-side modulation DNN (e.g., DNN 434) and a coding module (e.g., coding module 428) using a static algorithm. Each UE in the group of UEs processes downlink communications using a respective hybrid receiver processing chain (e.g., a respective instance of hybrid receiver processing chain 426) including a respective UE-side demodulation DNN (e.g., demodulation DNN 442) and a respective decoding module (e.g., decoding module 444) using a static algorithm, with each UE forming a respective UE-side demodulation DNN using the common ML configuration shown at 715. Alternatively or additionally, each UE in the group of UEs processes uplink communications using a hybrid transmitter processing chain (e.g., hybrid transmitter processing chain 424) including a respective UE-side modulation DNN (e.g., modulation DNN 434) and a respective encoding module (e.g., encoding module 428) using a static algorithm. Alternatively or additionally, base station 120 processes uplink communications using a hybrid receiver processing chain (e.g., hybrid receiver processing chain 426) including a BS-side demodulation DNN (e.g., demodulation DNN 442) and a decoding module (e.g., decoding module 444) using a static algorithm.

[0095] At 730, at least one UE 110 included in the group of UEs detects a training condition. Illustratively, the UE 110 detects the occurrence / reoccurrence of the training time from a periodic training schedule. Alternatively or additionally, the UE detects "N" CRC failures as described at 565 of FIG. 5, detects signal quality measurements and / or link quality measurements that do not meet a performance threshold, receives feedback from the base station 120, etc. Accordingly, and in response to detecting the training condition, the UE 110 trains a UE-side DNN at 735, such as that described at 565 of FIG. 5 and / or 655 of FIG. 6. At 740, each UE 110 transmits an ML configuration update to the base station 120 as described at 575 of FIG. 5 and / or 665 of FIG. 6. For visual clarity, diagram 700 illustrates each UE 110 in a group of UEs simultaneously detecting training conditions, performing training of the UE-side DNN, and transmitting an ML configuration update to base station 120, although in other aspects, each UE detects its respective training conditions and performs training at different times (e.g., asynchronously) from each other.

[0096] At 745, the base station 120 uses the received ML configuration updates from each UE in the group of UEs and a federated learning technique to identify one or more updated ML configurations. For example, the base station 120 applies a federated learning technique that aggregates updated ML configurations received from multiple UEs (e.g., the updated ML configuration transmitted at 740), possibly without revealing private data used at the UEs, to generate the updated ML configuration. Illustratively, the base station 120 performs a weighted average that aggregates ML parameters, gradients, etc. As another example, each UE 110 reports a gradient result indicating an optimal ML configuration based on function processing cost (e.g., processing time, processing accuracy) based on its own individual input data, and the base station 120 averages the gradients. In some aspects, multiple devices report learned ML architecture updates and / or changes from the initial and / or common ML configuration. The updated ML configuration may correspond to a UE-side demodulation DNN and / or a UE-side modulation DNN. In some aspects, the base station 120 additionally determines updates to the BS-side modulation DNN and / or the BS-side demodulation DNN as described with reference to FIGS.

[0097] At 750, base station 120 indicates the updated common ML configuration to at least some of the UEs in the group of UEs. This may include indicating the updated common ML configuration using DCI, using CSI-RS, pilot signals, etc.

[0098] At 755, at least some of the UEs in the group of UEs update their respective UE-side DNNs using the updated ML configuration shown at 750. At 760, processing proceeds to signaling and data transactions as performed at 725, with each UE 110 then using the updated UE-side DNN to process communications, uplink, and / or downlink.

[0099] At 765, the base station 120 optionally updates one or more BS-side DNNs using the updated ML configuration for the BS-side DNN, shown by dashed lines in diagram 700. At 770, the base station 120 uses the updated BS-side DNN to process communications, uplink, and / or downlink.

[0100] Example Method Example methods 800, 900, 1000, and 1100 are described with reference to FIGS. 8-11 in accordance with one or more aspects of a hybrid wireless communication processing chain including DNN and static algorithm modules.

[0101] 8 illustrates an example method 800 used to implement aspects of a hybrid wireless communication processing chain including a DNN and a static algorithm module. For example, in aspects of method 800, a first wireless communication device communicates with a second wireless communication device using a hybrid transmitter processing chain. In some implementations, the first wireless communication device is a base station (e.g., base station 120) and the second wireless communication device is a UE (e.g., UE 110). In other implementations, the first wireless communication device is a UE (e.g., UE 110) and the second wireless communication device is a base station (e.g., base station 120).

[0102] At 805, the first wireless communication device selects a modulation machine learning configuration (ML configuration) to form a modulation deep neural network (modulation DNN) that generates a modulated signal using the coded bits as input. As one example, a base station (e.g., base station 120) selects the BS-side modulation ML configuration as illustrated in 505 of FIG. 5. As another example, a UE (e.g., UE 110) selects the UE-side modulation ML configuration as illustrated in 615 of FIG. 6.

[0103] At 810, the first wireless communication device forms, based on the modulation ML configuration, a modulation DNN as part of a hybrid transmitter processing chain including the modulation DNN and at least one static algorithm module. Illustratively, a base station (e.g., base station 120) forms the BS-side modulation DNN (e.g., modulation DNN 434) as part of the hybrid transmitter processing chain (e.g., hybrid transmitter processing chain 424) as described at 525 of FIG. 5 and with reference to FIG. 4. Alternatively, a UE (e.g., UE 110) forms the UE-side modulation DNN (e.g., modulation DNN 434) as part of the hybrid transmitter processing chain (e.g., hybrid transmitter processing chain 424) as described at 625 of FIG. 6 and with reference to FIG. 4.

[0104] At 815, the first wireless communication device processes wireless communications associated with the second wireless communication device using a hybrid transmitter processing chain. As one example, a base station (e.g., base station 120) processes downlink communications directed to a UE (e.g., UE 110) using a hybrid transmitter processing chain (e.g., hybrid transmitter processing chain 424) as described at 535 of FIG. 5 and with reference to FIG. 4. As another example, a UE (e.g., UE 110) processes uplink communications directed to a base station (e.g., base station 120) using a hybrid transmitter processing chain (e.g., hybrid transmitter processing chain 424) as described at 635 of FIG. 6 and with reference to FIG. 4.

[0105] 9 illustrates an example method 900 used to implement aspects of a hybrid wireless communication processing chain including a DNN and a static algorithm module. For example, in aspects of method 900, a first wireless communication device communicates with a second wireless communication device using a hybrid receiver processing chain. In some implementations, the first wireless communication device is a base station (e.g., base station 120) and the second wireless communication device is a UE (e.g., UE 110). In other implementations, the first wireless communication device is a UE (e.g., UE 110) and the second wireless communication device is a base station (e.g., base station 120).

[0106] At 905, the first wireless communication device selects a demodulation machine learning configuration (ML configuration) to form a demodulation deep neural network (demodulation DNN) that uses the modulated signal as an input to generate coded bits. As one example, a base station (e.g., base station 120) selects the BS-side demodulation ML configuration as illustrated at 630 in FIG. 6. As another example, a UE (e.g., UE 110) selects the UE-side demodulation ML configuration as illustrated at 515 in FIG. 5.

[0107] At 910, the first wireless communication device forms the demodulation DNN as part of a hybrid receiver processing chain including the demodulation DNN and at least one static algorithm module based on the demodulation ML configuration. Illustratively, the base station (e.g., base station 120) forms the BS-side demodulation DNN (e.g., demodulation DNN 442) as a hybrid receiver processing chain, as described in 630 of FIG. 6 and with reference to FIG. 4. Receiving 5 and 6. Alternatively, the UE (e.g., UE 110) forms a UE-side demodulation DNN (e.g., demodulation DNN 442) as part of the hybrid receiver processing chain (e.g., hybrid receiver processing chain 426) as described in 530 of FIG. 5 and with reference to FIG. 4.

[0108] At 915, the first wireless communication device processes wireless communications associated with the second wireless communication device using a hybrid receiver processing chain. As one example, a base station (e.g., base station 120) processes uplink communications from a UE (e.g., UE 110) using a hybrid receiver processing chain (e.g., hybrid receiver processing chain 426) as described at 645 of FIG. 6 and with reference to FIG. 4. As another example, a UE (e.g., UE 110) processes downlink communications from a base station (e.g., base station 120) using a hybrid receiver processing chain (e.g., hybrid receiver processing chain 426) as described at 545 of FIG. 5 and with reference to FIG. 4.

[0109] 10 illustrates an example method 1000 used to implement aspects of a hybrid wireless communication processing chain including DNN and static algorithm modules. In some implementations, the operations of method 1000 are performed by a base station, such as base station 120.

[0110] At 1005, the base station selects a machine learning configuration (ML configuration) to form a DNN that (i) generates a modulated downlink signal using coded bits as input, or (ii) generates coded bits using a modulated uplink signal as input. For example, the base station (e.g., base station 120) selects a BS-side modulation ML configuration for a BS-side modulation DNN (e.g., modulation DNN 434) that processes downlink communications as described at 505 of FIG. 5. As another example, the base station (e.g., base station 120) selects a UE-side demodulation configuration for a UE-side demodulation DNN (e.g., demodulation DNN 442). In some aspects, the base station selects a UE-side modulation ML configuration for the UE as the ML configuration as described at 605 of FIG. 6 and with reference to FIG. 4.

[0111] At 1010, the base station indicates the ML configuration to the UE. Illustratively, the base station (e.g., base station 120) indicates the BS-side modulation ML configuration and / or the UE-side demodulation ML configuration to the UE (e.g., UE 110) in a DCI field or using a reference signal, as described in 510 of FIG. 5 and with reference to FIG. 4. In some aspects, the base station (e.g., base station 120) indicates the UE-side modulation ML configuration, as described in 610 of FIG. 6 and with reference to FIG. 5.

[0112] At 1015, the base station forms a base station-side DNN included in a hybrid wireless communication processing chain including a base station-side DNN and at least one static algorithm module based on the indicated ML configuration. Illustratively, when the base station selects and / or indicates a BS-side modulation ML configuration at 1005 and 1010, the base station (e.g., base station 120) forms a BS-side modulation DNN (e.g., modulation DNN 432) included in a hybrid transmitter processing chain (e.g., hybrid transmitter processing chain 424) as described at 525 of FIG. 5 and with reference to FIG. 4. Alternatively or additionally, when the base station selects and / or indicates a UE-side demodulation ML configuration at 1005 and 1010, the base station (e.g., base station 120) forms a BS-side modulation DNN having a complementary BS-side ML configuration. In some aspects, such as when the base station selects and indicates a UE-side modulation ML configuration (e.g., as described in 605 and 610 of FIG. 6), the base station (e.g., base station 120) forms a BS-side demodulation DNN (e.g., demodulation DNN 442) included in a hybrid receiver processing chain (e.g., hybrid receiver processing chain 426), as described with reference to FIG. 4.

[0113] At 1020, the base station processes wireless communications associated with the UE using a hybrid wireless communication processing chain. Illustratively, the base station (e.g., base station 120) processes downlink communications directed to the UE (e.g., UE 110) using a BS-side modulation DNN (e.g., modulation DNN 434) included in a hybrid transmitter processing chain (e.g., hybrid transmitter processing chain 424), as illustrated at 535 in FIG. 5. Alternatively or additionally, the base station (e.g., base station 120) processes uplink communications received from the UE (e.g., UE 110) using a BS-side demodulation DNN (e.g., demodulation DNN 442) included in a receiver processing chain (e.g., hybrid receiver processing chain 426), as illustrated at 645 in FIG.

[0114] 11 illustrates an example method 1100 used to implement aspects of a hybrid wireless communication processing chain including DNN and static algorithm modules. In some implementations, the operations of the method 1100 are performed by a user equipment, such as a UE 110.

[0115] At 1105, the UE receives from the base station an indication of an ML configuration to form a DNN to process wireless communications associated with the base station. As one example, the UE (e.g., UE 110) receives an indication of a BS-side modulation ML configuration, as shown at 510 in Figure 5. As another example, the UE (e.g., UE 110) receives an indication of a UE-side ML configuration, such as by receiving an indication of a UE-side modulation ML configuration as illustrated at 610 in Figure 6 and / or by receiving an indication of a UE-side demodulation ML configuration as illustrated at 510 in Figure 5.

[0116] At 1110, the UE selects, based on the indicated ML configuration, a UE-side ML configuration that forms a UE-side DNN that (i) uses the modulated downlink signal as an input and generates coded bits as an output, or (ii) uses the coded bits as an input and generates a modulated uplink signal. Illustratively, the UE (e.g., UE 110) selects a UE-side demodulation ML configuration as illustrated at 515 in FIG. 5 and / or selects a UE-side modulation ML configuration as illustrated at 615 in FIG. 6.

[0117] At 1115, the UE uses the UE-side ML configuration to form a UE-side DNN as part of a hybrid wireless communication processing chain that includes at least one static algorithm module and a UE-side DNN. This may include the UE (e.g., the UE 110) forming a UE-side demodulation DNN included in a hybrid receiver processing chain (e.g., the hybrid receiver processing chain 426) as described at 530 of FIG. 5 and with reference to FIG. 4, or the UE (e.g., the UE 110) forming a UE-side modulation DNN included in a hybrid transmitter processing chain (e.g., the hybrid transmitter processing chain 424) as described at 625 of FIG. 6 and with reference to FIG. 4.

[0118] At 1120, the UE processes wireless communications associated with a base station using a hybrid wireless communication processing chain. Illustratively, the UE (e.g., UE 110) processes downlink communications from a base station (e.g., base station 120) using a UE-side demodulation DNN (e.g., demodulation DNN 442) included in a receiver processing chain (e.g., hybrid receiver processing chain 426), as illustrated at 545 in FIG. 5. Alternatively or additionally, the user equipment (e.g., UE 110) processes uplink communications directed to a base station (e.g., base station 120) using a UE-side modulation DNN (e.g., modulation DNN 434) included in a transmitter processing chain (e.g., hybrid transmitter processing chain 424), as illustrated at 635 in FIG.

[0119] The order in which the method blocks of methods 800-1100 are described should not be construed as a limitation, and any number of the described method blocks may be skipped or combined in any order to implement the method or alternative methods. Generally, any of the components, modules, methods, and operations described herein may be implemented using software, firmware, hardware (e.g., fixed logic circuitry), manual processing, or any combination thereof. Some operations of the example methods may be described in the general context of executable instructions stored in computer-readable storage memory that is local and / or remote to a computer processing system, and implementations may include software applications, programs, functions, and the like. Alternatively, or additionally, any of the functionality described herein may be implemented at least in part by one or more hardware logic components, such as, without limitation, field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-chip (SoCs), combinatorial programmable logic devices (CPLDs), and the like.

[0120] Although the techniques and devices for a hybrid wireless communications processing chain including a DNN and a static algorithm module have been described in feature- and / or method-specific language, it should be understood that the subject matter of the appended claims is not necessarily limited to the particular features or methods described. Rather, the particular features and methods are disclosed as example implementations of a hybrid wireless communications processing chain including a DNN and a static algorithm module.

[0121] Example Below, some examples of the subject matter described herein are described.

[0122] In one example, a method is implemented by a first wireless communication device to communicate with a second wireless communication device using a hybrid wireless communication processing chain. The method includes selecting, using the first wireless communication device, a modulation machine learning (ML) configuration to form a modulation deep neural network (DNN) that generates a modulated signal using coded bits received from a coding module as inputs, forming the modulation DNN as part of a hybrid transmitter processing chain that includes the modulation DNN and at least one static algorithm module based on the modulation ML configuration, and transmitting a wireless communication associated with the second wireless communication device using the hybrid transmitter processing chain.

[0123] Processing wireless communications associated with the second wireless communication device using the hybrid transmitter processing chain may optionally include transmitting the modulated signal to the second wireless communication device.

[0124] Selecting a modulation ML configuration optionally further includes selecting a modulation ML configuration that forms a DNN that performs multiple-input multiple-output (MIMO) antenna processing. The at least one static algorithm module may optionally be an encoding module. The method may optionally further include generating coded bits using the encoding module. Generating coded bits may optionally further include the encoding module using one or more of a low-density parity-check (LPDC) coding algorithm, a polar coding algorithm, a turbo coding algorithm, or a Viterbi coding algorithm.

[0125] Selecting the modulated ML configuration may optionally include selecting a convolutional neural network architecture, a recurrent neural network architecture, a fully connected neural network architecture, or a partially connected neural network architecture.

[0126] The method may optionally further include indicating the modulated ML configuration to a second wireless communication device.

[0127] The first wireless communication device may be a base station. The second wireless communication device may be a user equipment (UE). Selecting the modulation ML configuration may optionally further include selecting a base station side (BS side) modulation ML configuration as the modulation DNN to form a BS side modulation DNN that generates a modulated downlink signal using the coded bits received from the coding module as input. Forming the modulation DNN may optionally further include forming a BS side modulation DNN. The method may optionally further include indicating the BS side modulation ML configuration to the UE. Indicating the BS side modulation ML configuration to the UE may further include indicating the BS side modulation ML configuration using a field in downlink control information (DCI) or transmitting a reference signal mapped to the BS side modulation ML configuration.

[0128] The method may optionally further include receiving hybrid automatic repeat request (HARQ) feedback from the UE and training a BS-side modulation DNN using the HARQ feedback. The method may optionally further include selecting a user equipment side (UE-side) modulation ML configuration that forms a UE-side modulation DNN for generating a modulated uplink signal and indicating the UE-side modulation ML configuration to the UE. The demodulation BS-side DNN may optionally be formed based on the UE-side modulation ML configuration. Indicating the UE-side modulation ML configuration to the UE may further include indicating the UE-side modulation ML configuration to the UE using downlink control information (DCI). The BS-side modulation ML configuration may optionally be a first BS-side ML configuration. The method may optionally further include receiving an indication of a user equipment-selected (UE-selected) UE-side demodulation ML configuration from the UE. The BS-side modulation DNN may optionally be updated using a second BS-side modulation ML configuration that is complementary to the UE-selected UE-side demodulation ML configuration. Receiving an indication of the UE-selected UE-side demodulation ML configuration may optionally further include receiving an indication of the UE-selected UE-side demodulation ML configuration in channel state information (CSI). The UE may optionally be a first UE. The first UE-side ML configuration update for the common ML configuration may optionally be received from the first UE. The common ML configuration may optionally be a demodulation ML configuration or a modulation ML configuration. The second UE-side ML configuration update for the common ML configuration may optionally be received from the second UE. The updated common ML configuration may optionally be selected using a federated learning technique, the first UE-side ML configuration update, and the second UE-side ML configuration update. The first UE and the second UE may optionally be instructed to update their respective UE-side DNNs using the updated common ML configuration.

[0129] The first wireless communication device may optionally be a user equipment (UE). The second wireless communication device may optionally be a base station. Selecting a modulation ML configuration may further include selecting a UE-side modulation ML configuration to form a UE-side modulation DNN that uses the coded bits as input to generate a modulated uplink signal. The at least one static algorithm module may optionally be a coding module. Transmitting a wireless communication may further include receiving the coded bits as input from the coding module and generating a modulated uplink signal using the UE-side modulation DNN in a hybrid transmitter processing chain and based on the coded bits. Selecting a modulation ML configuration may optionally further include receiving an indication of the UE-side modulation ML configuration from the base station and selecting the modulation ML configuration using the indication. Receiving the indication may optionally further include receiving the indication in a field of downlink control information (DCI) for a physical uplink shared channel (PUSCH). Selecting a UE-side modulation ML configuration may optionally further include selecting a UE-side modulation ML configuration from a predefined set of modulation ML configurations.

[0130] In another example, a method is implemented by a first wireless communication device to communicate with a second wireless communication device using a hybrid receiver processing chain, the method including selecting a demodulation machine learning (ML) configuration to form a demodulation deep neural network (DNN) using the modulated signal as an input and generating coded bits as an output, using the demodulation ML configuration to form the demodulation DNN as part of a hybrid receiver processing chain that includes at least one static algorithm module and the demodulation DNN, and receiving a wireless signal from the second wireless communication device using the hybrid receiver processing chain.

[0131] The at least one static algorithm module may optionally include a decoding module. The method may optionally further include generating decoded bits using the decoding module. Generating the decoded bits may optionally further include the decoding module using one or more of a low-density parity check (LPDC) decoding algorithm, a polar decoding algorithm, a turbo decoding algorithm, or a Viterbi decoding algorithm. Selecting a demodulation ML configuration may optionally further include selecting an ML configuration that forms a demodulation DNN to receive the modulated signal as a first input and decoding feedback from the decoding module as a second input. The method may optionally further include forming the demodulation DNN to receive one or more log-likelihood ratios from the decoding module as second inputs to the demodulation DNN. The method may optionally further include measuring a cost function of the demodulation DNN using at least one of a block error rate or a bit error rate. Optionally, it may be determined using the cost function that the performance of the demodulation DNN has deteriorated below a threshold. Based on determining the performance as degraded below a threshold, a training procedure for the demodulation DNN can be optionally initiated. The method may optionally further include determining to initiate a training procedure for the demodulation DNN based on analyzing one or more signal quality measurements or link quality measurements, or analyzing a cyclic redundancy check (CRC) for the recovered bits jointly generated by the demodulation DNN and the decoding module. The method may optionally further include identifying that the CRC for the recovered bits fails a predetermined number of consecutive times, and training the demodulation DNN based on the CRC failing the predetermined number of times.

[0132] The first wireless communication device may optionally be a user equipment (UE). The second wireless communication device may optionally be a base station. Selecting a demodulation ML configuration may optionally further include selecting a user equipment side (UE side) demodulation ML configuration that forms a UE side demodulation DNN as a demodulation DNN. Selecting a demodulation ML configuration may optionally further include receiving an indication of a base station side demodulation ML configuration from a base station and selecting a user equipment side (UE side) demodulation ML configuration using the base station side demodulation ML configuration. Receiving an indication from the base station may further include receiving the indication in downlink control information (DCI) or as a reference signal mapped to the base station side demodulation ML configuration. Selecting a UE side demodulation ML configuration may optionally further include selecting a first demodulation ML configuration based on the base station side modulation ML configuration indicated by the base station, determining that the demodulation DNN formed using the first demodulation ML configuration cannot meet a performance threshold, and selecting a second demodulation ML configuration that meets the performance threshold. The second demodulation ML configuration may optionally be indicated to the base station. Indicating the second demodulation ML configuration to the base station may optionally further include transmitting a sounding reference signal (SRS) mapped to the second demodulation ML configuration.

[0133] The first wireless communication device may optionally be a base station. The second wireless communication device may optionally be a user equipment (UE). Selecting a demodulation ML configuration may optionally further include selecting a base station side (BS side) demodulation ML configuration that forms a BS side demodulation deep neural network (DNN) using the modulated uplink signal as an input to generate decoded bits. Selecting the BS side demodulation ML configuration may optionally further include selecting the BS side demodulation ML configuration as a complementary ML configuration to the UE side demodulation ML configuration indicated to the UE.

[0134] As another example, an apparatus includes a wireless transceiver, a processor, and a computer-readable storage medium containing instructions that, when executed by the processor, direct the apparatus to perform any of the methods described herein.

[0135] As another example, a computer-readable storage medium contains instructions that, upon execution by a processor, direct an apparatus to perform any of the methods described herein.

Claims

1. 1. A method implemented by a base station (BS) for communicating with a first user equipment (UE) using a hybrid wireless communication processing chain, comprising: selecting, using the BS, a base station-side (BS-side) modulation machine learning (ML) configuration to form a BS-side modulation deep neural network (DNN) that generates a modulated signal using the coded bits received from the coding module as input; forming the BS-side modulation DNN based on the BS-side modulation ML configuration as part of a hybrid transmitter processing chain including the BS-side modulation DNN and at least one static algorithm module; transmitting a wireless communication associated with the first UE using the hybrid transmitter processing chain; and receiving a first UE-side ML configuration update for a common ML configuration from the first UE, wherein the common ML configuration is a demodulation ML configuration or a modulation ML configuration, and the method further comprises: receiving a second UE-side ML configuration update for the common ML configuration from a second UE; selecting an updated common ML configuration using a federated learning technique, the first UE-side ML configuration update, and the second UE-side ML configuration update; instructing the first UE and the second UE to update their respective UE-side DNNs using the updated common ML configuration.

2. Selecting the modulation ML configuration comprises: The method of claim 1 , further comprising selecting a modulation ML configuration to form a DNN that implements multiple-input multiple-output (MIMO) antenna processing.

3. The at least one static algorithm module is the encoding module, and the method comprises: The method of claim 1 , further comprising generating the coded bits using the coding module.

4. generating the coded bits the encoding module: Low-density parity-check (LPDC) coding algorithm, Polar coding algorithm, Turbo coding algorithm, or The method of claim 3 further comprising using one or more of the Viterbi encoding algorithms.

5. Selecting the modulation ML configuration comprises: Convolutional Neural Network Architecture, Recurrent Neural Network Architecture, fully connected neural network architecture, or The method of claim 1 , comprising selecting a partially connected neural network architecture.

6. The method of claim 1 , further comprising indicating the modulation ML configuration to the first UE.

7. The method of claim 1 , further comprising indicating the BS-side modulation ML configuration to the first UE.

8. Indicating the BS side modulation ML configuration to the first UE comprises: indicating the BS-side modulation ML configuration using a field in Downlink Control Information (DCI); or The method of claim 7 , further comprising transmitting a reference signal that is mapped to the BS-side modulation ML configuration.

9. receiving hybrid automatic repeat request (HARQ) feedback from the first UE; The method of claim 1 , further comprising: training the BS-side modulated DNN using the HARQ feedback.

10. selecting a user equipment side (UE side) modulation ML configuration forming a UE side modulation DNN for generating a modulated uplink signal; indicating the UE-side modulation ML configuration to the first UE; The method of claim 1 , optionally wherein selecting the UE-side modulation ML configuration further comprises selecting the UE-side modulation ML configuration from a predefined set of modulation ML configurations.

11. 11. The method of claim 10, wherein indicating the UE-side modulation ML configuration to the first UE further comprises indicating the UE-side modulation ML configuration to the first UE using downlink control information (DCI).

12. the BS-side modulation ML configuration is a first BS-side ML configuration, and the method includes: receiving a user equipment selected (UE selected) UE-side demodulation ML configuration indication from the first UE; 2. The method of claim 1, further comprising: updating the BS-side modulation DNN using a second BS-side modulation ML configuration that is complementary to the UE-selected UE-side demodulation ML configuration.

13. 13. The method of claim 12, wherein receiving the indication of the UE-selected UE-side demodulation ML configuration further comprises receiving the indication of the UE-selected UE-side demodulation ML configuration in channel state information (CSI).

14. 1. An apparatus comprising: A wireless transmitter / receiver; a processor; and a computer-readable storage medium comprising instructions that, when executed by the processor, direct the apparatus to perform the method of any one of claims 1 to 13.

15. A computer program product which, when executed by a processor, causes an apparatus to perform the method of any one of claims 1 to 13.

Images