PAPR reduction method, device and medium based on peak-reduction tone using neural network

Abstract

Various embodiments may include a method for reducing a peak-to-average power ratio of a wireless transmission waveform, performed in a transmitter circuit for wireless communication. Various embodiments may include receiving frequency-domain data tones; transforming the frequency-domain data tones into time-domain data signals; generating a time-domain PRT using the time-domain data signals using a set of peak-reduced tone (PRT) neural networks, wherein the set of PRT neural networks has been trained in conjunction with a boosting neural network and a receiver neural network; generating an output of the boosting neural network based on an input including a final combined time-domain signal including the time-domain PRT combined with a previous combined time-domain signal; and generating a time-domain wireless transmission waveform including the output of the boosting neural network combined with the final combined time-domain signal.

Description

[0001] Related applications

[0002] This application claims the benefit of priority of U.S. Provisional Patent Application No. 63 / 070,119, filed on August 25, 2020, entitled “NeuralAugmentation For Device Nonlinearity Mitigation In X-Node Machine Learning,” and U.S. Patent Application No. 17 / 394,928, filed on August 5, 2021, entitled “NeuralAugmentation For Device Nonlinearity Mitigation In X-Node Machine Learning,” the entire contents of which are incorporated herein by reference. Background Art

[0003] In wireless communication systems, such as those specified in standards for fifth-generation (5G) New Radio (NR), conventional systems using orthogonal frequency division multiplexing (OFDM) waveforms suffer from large peak-to-average power ratios (PAPRs). Mitigating large PAPRs can require significant power amplifier back-off at the expense of degraded power amplifier efficiency. Summary of the Invention

[0004] Various aspects include systems and methods implemented in transmitter circuitry of wireless communication devices, such as base stations and mobile wireless devices, for reducing the peak-to-average power ratio of a wireless transmission waveform. Aspects may include: receiving frequency-domain data tones; transforming the frequency-domain data tones into time-domain data signals; using the time-domain data signals to generate a time-domain PRT using a set of peak-reduced tone (PRT) neural networks, wherein the set of PRT neural networks has been trained in conjunction with a boosting neural network and a receiver neural network; generating an output of the boosting neural network based on an input comprising a final combined time-domain signal of the time-domain PRT combined with a previous combined time-domain signal; and generating a time-domain wireless transmission waveform comprising the output of the boosting neural network combined with the final combined time-domain signal.

[0005] Some aspects may also include: generating a real feature map of the final combined time domain signal and a real feature map of the function of the final combined time domain signal, wherein generating the output of the enhanced neural network based on the input of the final combined time domain signal may include: generating the output of the enhanced neural network based on the input of the real feature map of the final combined time domain signal and the real feature map of the function of the final combined time domain signal.

[0006] Some aspects may also include: generating a complex feature map of the final combined time domain signal and a complex feature map of the function of the final combined time domain signal, wherein generating the output of the enhanced neural network based on the input of the final combined time domain signal includes: generating the output of the enhanced neural network based on the input of the complex feature map of the final combined time domain signal and the complex feature map of the function of the final combined time domain signal.

[0007] Some aspects may also include training the enhanced neural network using a weighted average of a first error and a second error, wherein the first error is a reconstruction from the frequency domain data tones and a comparison of the frequency domain data tones, and wherein the second error is a distortion function output based on energy outside of the allocated frequency band.

[0008] Some aspects may also include training the PRT neural network set using the weighted average of the first error and the second error.

[0009] Some aspects may also include determining whether the wireless communication device is configured to implement the enhanced neural network; and in response to determining that the wireless communication device is configured to implement the enhanced neural network, selecting a neural network implementation from a group of neural network implementations based on a modulation and coding scheme.

[0010] Some aspects may also include transmitting the time-domain wireless transmission waveform to another wireless communication device having the receiver neural network.

[0011] Some aspects may also include: receiving a neural network indicator from another wireless communication device having the receiver neural network, the neural network indicator configured to indicate to the wireless communication device the enhanced neural network that has been trained with the set of PRT neural networks and the receiver neural network; and selecting the enhanced neural network from a plurality of enhanced neural networks based on the neural network indicator.

[0012] Some aspects may also include receiving weights for the enhanced neural network from the other wireless communication device, wherein generating the time-domain wireless transmission waveform comprising the output of the enhanced neural network combined with the final combined time-domain signal may include generating the time-domain wireless transmission waveform using the weights for the enhanced neural network.

[0013] A further aspect may include a wireless communication device having transmit circuitry and / or receive circuitry configured to perform the operations of any of the methods outlined above. A further aspect may include a wireless computing device having means for performing the functions of any of the methods outlined above. A further aspect may include a non-transitory processor-readable medium having processor-executable instructions stored thereon, the processor-executable instructions configured to cause a processor of the wireless computing device to perform the operations of any of the methods outlined above. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the claims and, together with the general description given above and the detailed description given below, serve to explain features of the claims.

[0015] Figure 1 is a system block diagram illustrating an example of a communication system suitable for implementing any of the various embodiments.

[0016] Figure 2 is a component block diagram illustrating an example computing and wireless modem system suitable for implementing any of the various embodiments.

[0017] Figure 3 is a component block diagram illustrating an example of a software architecture suitable for implementing any of the various embodiments, including a radio protocol stack for user plane and control plane in wireless communications.

[0018] Figure 4A and 4B is a component block diagram illustrating an example of a system configured for managing information transmission for wireless communications suitable for implementing any of the various embodiments.

[0019] Figure 5A and 5B is a component block diagram illustrating an example of an encoder circuit configured to implement machine learning and / or machine learning-based encoding for nonlinearity mitigation using peak reduced pitch (PRT) and neural enhancement, in accordance with various embodiments.

[0020] Figure 6A and 6B is a component block diagram illustrating an example of a neural network decoder configured to implement machine learning and / or machine learning-based decoding for nonlinearity mitigation using PRT and neural enhancement, in accordance with various embodiments.

[0021] Figure 7is a component signaling diagram illustrating an example of a system configured to implement machine learning-based encoding and decoding for nonlinearity mitigation using PRT and neural enhancement, in accordance with various embodiments.

[0022] Figure 8 is a component signaling diagram illustrating an example of a system configured to implement machine learning-based encoding and decoding for nonlinearity mitigation using PRT and neural enhancement, in accordance with various embodiments.

[0023] Figure 9 is a process flow diagram illustrating an example of a method for delivering machine learning-based data for nonlinearity mitigation using PRT and neural enhancement, in accordance with various embodiments.

[0024] Figure 10A and 10B is a process flow diagram illustrating an example method for implementing machine learning-based encoding and decoding for nonlinearity mitigation using PRT and processing of enhanced neural networks, according to various embodiments.

[0025] Figure 11 is a process flow diagram illustrating an example of a method for training a neural network for generating a time-domain PRT based on data tones at the transmitter side and reconstructing the data tones at the receiver, in accordance with various embodiments.

[0026] Figure 12 is a process flow diagram illustrating an example of a method for generating a transmit waveform from data tones using a neural network, in accordance with various embodiments.

[0027] Figure 13 is a process flow diagram illustrating an example of a method for reconstructing a time-domain data signal from a transmission waveform using a neural network, according to various embodiments.

[0028] Figure 14 is a process flow diagram illustrating an example of a method for generating enhanced neural network outputs according to various embodiments.

[0029] Figure 15 is a carrier signal block diagram illustrating an example of a transmission waveform with data tones and PRT on orthogonal subcarriers according to various embodiments.

[0030] Figure 16 is a component block diagram illustrating an example network computing device.

[0031] Figure 17 is a component block diagram illustrating an example wireless device.

[0032] Like reference numbers and designations in the various drawings indicate like elements. DETAILED DESCRIPTION

[0033] For the purpose of describing the innovative aspects of the present disclosure, the following description relates to certain implementations. However, it will be readily appreciated by those skilled in the art that the teachings herein can be applied in many different ways.

[0034] Various embodiments provide methods that can be implemented in wireless communication devices (e.g., base stations, wireless access points, mobile devices, etc.) that utilize a trained PRT neural network module on the transmitter side of a wireless communication link to determine a PRT appropriate for a given transmission, and utilize an associated trained receiver neural network module on the receiver side of the wireless communication link to remove the PRT from the received signal and extract information about target data tones carried by the PRT.

[0035] The described embodiments may be implemented in any device, system, or network capable of transmitting and receiving radio frequency (RF) signals in accordance with any of the Institute of Electrical and Electronics Engineers (IEEE) 16.11 standards, or any of the following: IEEE 802.11 standards, Standard, Code Division Multiple Access (CDMA), Frequency Division Multiple Access (FDMA), Time Division Multiple Access (TDMA), Global System for Mobile Communications (GSM), GSM / General Packet Radio Service (GPRS), Enhanced Data GSM Environment (EDGE), Terrestrial Trunked Radio (TETRA), Wideband-CDMA (W-CDMA), Evolution-Data Optimized (EV-DO), IxEV-DO, EV-DO Rev A, EV-DO Rev B, High-Speed ​​Packet Access (HSPA), High-Speed ​​Downlink Packet Access (HSDPA), High-Speed ​​Uplink Packet Access (HSUPA), Evolved High-Speed ​​Packet Access (HSPA+), Long Term Evolution (LTE), AMPS, Fifth Generation (5G) New Radio; or the device, system or network is capable of sending and receiving other signals for communication within a wireless, cellular or Internet of Things (IoT) network (such as a system utilizing 3G, 4G or 5G technology or another embodiment thereof).

[0036] The term "wireless device" is used herein to refer to any or all of the following: wireless router devices, wireless appliances, cellular phones, smartphones, portable computing devices, personal or mobile multimedia players, laptop computers, tablet computers, smartbooks, ultrabooks, handheld computers, wireless email receivers, multimedia Internet-enabled cellular phones, medical devices and equipment, biometric sensors / devices, wearable devices (including smart watches, smart clothing, smart glasses, smart wristbands, smart jewelry (such as smart rings, smart bracelets, etc.)), entertainment devices (such as wireless game controllers, music and video players, satellite radio equipment, etc.), wireless network-enabled Internet of Things (IoT) devices (including smart meters / sensors, industrial manufacturing equipment, large and small machinery and appliances for home or business use), wireless communication elements within autonomous and semi-autonomous vehicles, wireless devices attached to or incorporated into various mobile platforms, global positioning system devices, and similar electronic devices that include memory, wireless communication components and programmable processors.

[0037] The term "system on a chip" (SOC) is used herein to refer to a single integrated circuit (IC) chip that includes multiple resources or processors integrated on a single substrate. A single SOC may include circuits for digital, analog, mixed-signal, and radio frequency functions. A single SOC may also include any number of general-purpose or specialized processors (digital signal processors, modem processors, video processors, etc.), memory blocks (such as ROM, RAM, flash memory, etc.), and resources (such as timers, voltage regulators, oscillators, etc.). The SOC may also include software for controlling the integrated resources and processors and for controlling peripheral devices.

[0038] The term "system-in-package" (SIP) may be used herein to refer to a single module or package that contains multiple resources, computing units, cores, or processors on two or more IC chips, substrates, or SOCs. For example, a SIP may include a single substrate on which multiple IC chips or semiconductor dies are stacked in a vertical configuration. Similarly, a SIP may include one or more multi-chip modules (MCMs) on which multiple ICs or semiconductor dies are packaged into a unified substrate. A SIP may also include multiple independent SOCs that are coupled together via high-speed communication circuits and are tightly packaged, such as on a single motherboard or in a single wireless device. The proximity of the SOCs facilitates high-speed communication and the sharing of memory and resources.

[0039] The traditional orthogonal frequency division multiplexing (OFDM) waveform adopted in the 5G NR specification suffers from a large peak-to-average power ratio (PAPR). Without other mitigations of the waveform's PAPR, the transmitter may need to reduce power amplification (i.e., implement power amplifier back-off) at the expense of degraded power amplifier efficiency in order to avoid distortion caused by power amplifier nonlinearities. Signal processing methods for PAPR reduction include tone reservation schemes, in which peak reducing tones (PRTs) orthogonal to the data tones are used to shape the transmission waveform (i.e., the OFDM symbol containing the PRT and data tones) in the time domain. The PRT is designed to reduce the peak value of the amplitude of the transmission waveform in the time domain. However, when the PRT is found by traditional signal processing algorithms, there is no known relationship between the data tones and the PRT. In other words, the mapping between the data tones and the PRT can be arbitrary. Therefore, the receiver is independent of the content of the PRT, and the PRT only presents overhead to the receiver. When the PRT is paired with data tones, the peak power of the transmit waveform is reduced at the expense of increased average transmit power, and the error vector magnitude (EVM) is maintained for the data tones.

[0040] Embodiments described herein use a neural network trained via machine learning that is configured to map data tones to PRTs for transmission from a specific transmitter and / or for transmission to a specific receiver. Machine learning can be used to train a PRT neural network and a receiver neural network pair. In some embodiments, the PRT neural network can be trained to pair data tones with PRTs based on an input of a data signal in the time domain derived from the data tones in the frequency domain. The receiver neural network can be trained to demodulate a transmission waveform of an OFDM symbol in the time domain having data tones and PRTs to generate a reconstruction of the data tones in the frequency domain based on the input of the transmission waveform. In some embodiments, the PRT neural network and the receiver neural network can be trained for a specific transmitter, such as based on the hardware configuration of the transmitter. In some embodiments, the PRT neural network and the receiver neural network can be trained for a specific receiver, such as based on the hardware configuration of the receiver.

[0041] In some embodiments, a transmitter and a receiver can share configurations of a PRT neural network and / or a receiver neural network. For example, a transmitter (such as a wireless device) can share with a receiver (such as a base station or node) the configuration of a receiver neural network implemented by the receiver, which can be a receiver neural network trained in conjunction with a PRT neural network and an augmented neural network implemented by the transmitter. As a further example, a receiver can share with a transmitter the configuration of a PRT neural network and an augmented neural network implemented by the transmitter, which can be a PRT neural network and an augmented neural network trained in conjunction with a receiver neural network implemented by the receiver.

[0042] In some embodiments, the transmitter and receiver may be pre-configured with multiple PRT neural networks and / or receiver neural networks, and the configuration of the shared PRT neural network and / or receiver neural network may include an indicator of the configuration of the shared PRT neural network and / or receiver neural network. The transmitter and / or receiver may use the indicator to select the configuration of the PRT neural network and / or receiver neural network.

[0043] In some embodiments, sharing the configuration of the PRT neural network and / or the receiver neural network can include sharing weights resulting from training the PRT neural network and / or the receiver neural network. Sharing the configuration of the PRT neural network and / or the receiver neural network can include sending indicators and / or weights to the transmitter and / or receiver.

[0044] In some embodiments, a transmitter can use a default PRT neural network and / or a PRT neural network selected in response to receiving an indicator of a configuration of the PRT neural network to generate a PRT for a data tone. The transmitter can generate a transmission waveform by combining the data tone and the PRT. The transmitter can send the transmission waveform to a receiver. The receiver can receive the transmission waveform and use the default receiver neural network and / or the receiver neural network selected in response to receiving an indicator of a configuration of the receiver neural network to demodulate the transmission waveform to reconstruct the data tone. For the PRT neural network and the receiver neural network that are trained together, the PRT neural network used to generate the PRT to be combined with the data tone to generate the transmission waveform is known to the receiver implementing the receiver neural network. In this way, the PRT can contain information that is used to facilitate the reconstruction of the data tone. The information contained in the PRT can improve demodulation performance, such as improving the accuracy of the reconstruction of the data tone compared to transmitting a PRT generated by a traditional signal processing algorithm.

[0045] Figure 11 is a system block diagram illustrating an example communication system 100 suitable for implementing any of the various embodiments. The communication system 100 may be a 5G New Radio (NR) network, or any other suitable network, such as a Long Term Evolution (LTE) network.

[0046] The communication system 100 may include a heterogeneous network architecture including a core network 140 and various mobile devices (in Figure 1 120e). The communication system 100 may also include a plurality of base stations 110 (shown as BS 110a, BS 110b, BS 110c, and BS 110d) and other network entities. A base station is an entity that communicates with wireless devices (mobile devices) and may also be referred to as a NodeB, NodeB, LTE evolved NodeB (eNB), access point (AP), radio head, transmit receive point (TRP), new radio base station (NR BS), 5G NodeB (NB), next generation NodeB (gNB), etc. Each base station may provide communication coverage for a specific geographic area. In 3GPP, the term "cell" may refer to the coverage area of ​​a base station, a base station subsystem serving the coverage area, or a combination thereof, depending on the context in which the term is used.

[0047] The base stations 110a-110d may provide communication coverage for a macro cell, a pico cell, a femto cell, another type of cell, or a combination thereof. A macro cell may cover a relatively large geographic area (e.g., several kilometers in radius) and may allow unrestricted access by mobile devices with service subscription. A pico cell may cover a relatively small geographic area and may allow unrestricted access by mobile devices with service subscription. A femto cell may cover a relatively small geographic area (e.g., a residence) and may allow restricted access by mobile devices associated with the femto cell (e.g., mobile devices in a closed subscriber group (CSG)). A base station for a macro cell may be referred to as a macro BS. A base station for a pico cell may be referred to as a pico BS. A base station for a femto cell may be referred to as a femto BS or a home BS. In Figure 1 In the example shown in FIG, base station 110a may be a macro BS for macrocell 102a, base station 110b may be a pico BS for picocell 102b, and base station 110c may be a femto BS for femtocell 102c. Base stations 110a-110d may support one or more (e.g., three) cells. The terms "eNB," "base station," "NR BS," "gNB," "TRP," "AP," "Node B," "5G NB," and "cell" may be used interchangeably herein.

[0048] In some examples, the cells may not be stationary, and the geographic area of ​​the cells may move depending on the location of the mobile base station. In some examples, base stations 110a-110d may be interconnected with each other and with one or more other base stations or network nodes (not shown) in communication system 100 via various types of backhaul interfaces (such as direct physical connections, virtual networks, or combinations thereof) using any suitable transport network.

[0049] Base stations 110a-110d may communicate with core network 140 over wired or wireless communication links 126. Wireless devices 120a-120e may communicate over wireless communication links 122 with base stations 110a-110d.

[0050] The wired communication link 126 may use various wired networks (e.g., Ethernet, television cable, telephone, fiber optic, and other forms of physical network connections) that may use one or more wired communication protocols (such as Ethernet, Point-to-Point Protocol, High-Level Data Link Control (HDLC), High-Level Data Communications Control Protocol (ADCCP), and Transmission Control Protocol / Internet Protocol (TCP / IP)).

[0051] The communication system 100 may also include a relay station (e.g., relay BS 110d). A relay station is an entity that can receive data transmissions from an upstream station (e.g., a base station or a mobile device) and send the data to a downstream station (e.g., a wireless device or a base station). A relay station may also be a mobile device that can relay transmissions for other wireless devices. Figure 1 In the example shown in FIG, a relay station 110d can communicate with the macro base station 110a and the wireless device 120d to facilitate communication between the base station 110a and the wireless device 120d. A relay station may also be referred to as a relay base station, relay base station, relay, etc.

[0052] The communication system 100 may be a heterogeneous network including different types of base stations (e.g., macro base stations, pico base stations, femto base stations, relay base stations, etc.). These different types of base stations may have different transmit power levels, different coverage areas, and different impacts on interference in the communication system 100. For example, a macro base station may have a high transmit power level (e.g., 5 to 40 watts), while a pico base station, a femto base station, and a relay base station may have a lower transmit power level (e.g., 0.1 to 2 watts).

[0053] The network controller 130 may be coupled to a set of base stations and may provide coordination and control for these base stations. The network controller 130 may communicate with the base stations via a backhaul. The base stations may also communicate with each other directly or indirectly, for example, via a wireless or wired backhaul.

[0054] Wireless devices 120a, 120b, 120c may be dispersed throughout the communication system 100, and each wireless device may be stationary or mobile.A wireless device may also be referred to as an access terminal, terminal, mobile station, subscriber unit, station, etc.

[0055] The macro base station 110a may communicate with the communication network 140 over a wired or wireless communication link 126. The wireless devices 120a, 120b, 120c may communicate over a wireless communication link 122 with the base stations 110a-110d.

[0056] The wireless communication links 122, 124 may include multiple carrier signals, frequencies, or frequency bands, each of which may include multiple logical channels. The wireless communication links 122 and 124 may utilize one or more radio access technologies (RATs). Examples of RATs that may be used in the wireless communication links include 3GPP LTE, 3G, 4G, 5G (e.g., NR), GSM, Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Worldwide Interoperability for Microwave Access (WiMAX), Time Division Multiple Access (TDMA), and other mobile phone communication technology cellular RATs. Further examples of RATs that may be used in one or more of the various wireless communication links 122, 124 within the communication system 100 include medium-range protocols (such as Wi-Fi, LTE-U, LTE Direct, LAA, MuLTEfire) and relatively short-range RATs (such as ZigBee, Bluetooth, and Bluetooth Low Energy (LE)).

[0057] Some wireless networks (e.g., LTE) utilize orthogonal frequency division multiplexing (OFDM) on the downlink and single carrier frequency division multiplexing (SC-FDM) on the uplink. OFDM and SC-FDM divide the system bandwidth into multiple (K) orthogonal subcarriers, which are also commonly referred to as tones, bins, etc. Each subcarrier can be modulated with data. In general, modulation symbols are sent in the frequency domain with OFDM and in the time domain with SC-FDM. The spacing between adjacent subcarriers can be fixed, and the total number of subcarriers (K) can depend on the system bandwidth. For example, the subcarrier spacing can be 15 kHz and the minimum resource allocation (called a "resource block") can be 12 subcarriers (or 180 kHz). Therefore, for system bandwidths of 1.25, 2.5, 5, 10, or 20 megahertz (MHz), the nominal fast file transfer (FFT) size can be equal to 128, 256, 512, 1024, or 2048, respectively. The system bandwidth can also be divided into subbands. For example, a subband may cover 1.08 MHz (ie, 6 resource blocks), and there may be 1, 2, 4, 8, or 16 subbands for system bandwidths of 1.25, 2.5, 5, 10, or 20 MHz, respectively.

[0058] Although the description of some embodiments may use terminology and examples associated with LTE technology, the various embodiments may be applicable to other wireless communication systems, such as New Radio (NR) or 5G networks. NR can utilize OFDM with a cyclic prefix (CP) on the uplink (UL) and downlink (DL) and can include support for half-duplex operation using time division duplexing (TDD). A single component carrier bandwidth of 100 MHz can be supported. An NR resource block can span 12 subcarriers with a subcarrier bandwidth of 75 kHz in a duration of 0.1 milliseconds (ms). Each radio frame can consist of 50 subframes and have a length of 10 ms. Therefore, each subframe can have a length of 0.2 ms. Each subframe can indicate the link direction (i.e., DL or UL) used for data transmission, and the link direction for each subframe can be switched dynamically. Each subframe can include DL / UL data and DL / UL control data. Beamforming can be supported and the beam direction can be dynamically configured. Multiple-input multiple-output (MIMO) transmission with precoding can also be supported. MIMO configurations in the DL can support up to eight transmit antennas with multi-layer DL transmissions of up to eight streams and up to two streams per wireless device. Multi-layer transmissions with up to two streams per wireless device can be supported. Multiple cell aggregation with up to eight serving cells can be supported. Alternatively, NR can support different air interfaces besides the OFDM-based air interface.

[0059] Some mobile devices may be considered machine type communication (MTC) or evolved or enhanced machine type communication (eMTC) mobile devices. MTC and eMTC mobile devices include, for example, robots, drones, remote devices, sensors, meters, monitors, location tags, etc., which can communicate with a base station, another device (e.g., a remote device), or some other entity. A wireless node may provide a connection to or to a network (e.g., a wide area network such as the Internet or a cellular network), for example, via a wired or wireless communication link. Some mobile devices may be considered Internet of Things (IoT) devices or may be implemented as NB-IoT (narrowband Internet of Things) devices. The wireless devices 120a-120e may be included inside a housing that houses components of the wireless device (such as a processor component, a memory component, similar components, or a combination thereof).

[0060] In general, any number of communication systems and any number of wireless networks can be deployed in a given geographic area. Each communication system and wireless network can support a specific radio access technology (RAT) and can operate on one or more frequencies. RAT can also be referred to as radio technology, air interface, etc. Frequency can also be referred to as carrier, frequency channel, etc. Each frequency can support a single RAT in a given geographic area to avoid interference between communication systems of different RATs. In some cases, NR or 5G RAT networks can be deployed.

[0061] Figure 2 is a component block diagram illustrating an example computing and wireless modem system 200 suitable for implementing any of the various embodiments. The various embodiments may be implemented on a variety of single-processor and multi-processor computer systems including system-on-chip (SOC) or system-in-package (SIP).

[0062] refer to Figure 1 and 2 , the example computing system 200 shown (which may be a SIP in some embodiments) includes: two SOCs 202, 204, which are coupled to a clock 206; a voltage regulator 208; and a wireless transceiver 266, which is configured to send wireless communications to a wireless device (such as base station 110a) via an antenna (not shown) and receive wireless communications from it via the antenna. In some embodiments, the first SOC 202 operates as a central processing unit (CPU) of the wireless device, executing instructions of a software application by performing arithmetic, logic, control, and input / output (I / O) operations specified by the instructions. In some embodiments, the second SOC 204 can operate as a dedicated processing unit. For example, the second SOC 204 can operate as a dedicated 5G processing unit, which is responsible for managing communications of high capacity, high speed (e.g., 5 Gbps, etc.) and / or extremely high frequency short wavelength (e.g., 28 GHz millimeter wave spectrum, etc.).

[0063] The first SOC 202 may include a digital signal processor (DSP) 210, a modem processor 212, a graphics processor 214, an application processor 216, one or more coprocessors 218 (e.g., vector coprocessors) connected to one or more of these processors, memory 220, custom circuitry 222, system components and resources 224, an interconnect / bus module 226, one or more temperature sensors 230, a thermal management unit 232, and a thermal power envelope (TPE) component 234. The second SOC 204 may include a 5G modem processor 252, a power management unit 254, an interconnect / bus module 264, multiple mmWave transceivers 256, memory 258, and various additional processors 260 (such as an application processor, a packet processor, etc.).

[0064] Each processor 210, 212, 214, 216, 218, 252, 260 may include one or more cores, and each processor / core may operate independently of the other processors / cores. For example, the first SOC 202 may include a processor that executes a first type of operating system (e.g., FreeBSD, LINUX, OS X, etc.) and a processor that executes a second type of operating system (e.g., MICROSOFT WINDOWS 10). In addition, any or all of the processors 210, 212, 214, 216, 218, 252, 260 may be included as part of a processor cluster architecture (e.g., a synchronous processor cluster architecture, an asynchronous or heterogeneous processor cluster architecture, etc.).

[0065] The first SOC 202 and the second SOC 204 may include various system components, resources, and custom circuits for managing sensor data, analog-to-digital conversion, wireless data transmission, and for performing other specialized operations, such as decoding data packets and processing encoded audio and video signals for presentation in a web browser. For example, the system components and resources 224 of the first SOC 202 may include power amplifiers, voltage regulators, oscillators, phase-locked loops, peripheral bridges, data controllers, memory controllers, system controllers, access ports, timers, and other similar components for supporting processors and software clients running on wireless devices. The system components and resources 224 and / or custom circuits 222 may also include circuits for docking with peripheral devices (such as cameras, electronic displays, wireless communication devices, external memory chips, etc.).

[0066] The first SOC 202 and the second SOC 204 can communicate via an interconnect / bus module 250. The various processors 210, 212, 214, 216, 218 can be interconnected to one or more memory elements 220, system components and resources 224, and custom circuits 222, and a thermal management unit 232 via an interconnect / bus module 226. Similarly, the processor 252 can be interconnected to a power management unit 254, a millimeter wave transceiver 256, a memory 258, and various additional processors 260 via an interconnect / bus module 264. The interconnect / bus modules 226, 250, 264 can include an array of reconfigurable logic gates and / or implement a bus architecture (e.g., CoreConnect, AMBA, etc.). Communication can be provided through an advanced interconnect such as a high-performance network on chip (NoC).

[0067] The first SOC 202 and / or the second SOC 204 may also include input / output modules (not shown) for communicating with resources external to the SOC, such as a clock 206 and a voltage regulator 208. Resources external to the SOC (e.g., clock 206, voltage regulator 208) may be shared by two or more of the internal SOC processors / cores.

[0068] In addition to the example SIP 200 discussed above, various embodiments may be implemented in a wide variety of computing systems, which may include a single processor, multiple processors, multi-core processors, or any combination thereof.

[0069] Figure 3 is a component block diagram illustrating a software architecture 300 suitable for implementing any of the various embodiments, the software architecture 300 including radio protocol stacks for user plane and control plane in wireless communications. Figure 1-3 , a wireless device 320 may implement a software architecture 300 to facilitate communication between the wireless device 320 (e.g., wireless devices 120a-120e, 200) and a base station 350 (e.g., base station 110) of a communication system (e.g., 100). In various embodiments, the layers in the software architecture 300 may form logical connections with corresponding layers in the software of the base station 350. The software architecture 300 may be distributed across one or more processors (e.g., processors 212, 214, 216, 218, 252, 260). Although described with respect to a single radio protocol stack, in a multi-SIM (Subscriber Identity Module) wireless device, the software architecture 300 may include multiple protocol stacks, each of which may be associated with a different SIM (e.g., two protocol stacks associated with each of the two SIMs in a dual-SIM wireless communication device). Although described below with reference to LTE communication layers, the software architecture 300 may support any of a variety of standards and protocols for wireless communication and / or may include additional protocol stacks that support any of a variety of standards and protocols for wireless communication.

[0070] The software architecture 300 may include a non-access stratum (NAS) 302 and an access stratum (AS) 304. The NAS 302 may include functions and protocols for supporting packet filtering, security management, mobility control, session management, and services and signaling between a SIM of a wireless device (e.g., SIM 204) and its core network 140. The AS 304 may include functions and protocols for supporting communication between a SIM (e.g., SIM 204) and entities of a supported access network (e.g., base stations). Specifically, the AS 304 may include at least three layers (Layer 1, Layer 2, and Layer 3), each of which may include various sublayers.

[0071] In the user plane and control plane, layer 1 (L1) of the AS 304 may be a physical layer (PHY) 306, which may oversee functions that enable transmission or reception over the air interface via a wireless transceiver (e.g., 256). Examples of such physical layer 306 functions may include cyclic redundancy check (CRC) appending, coding blocks, scrambling and descrambling, modulation and demodulation, signal measurement, MIMO, etc. The physical layer may include various logical channels, including a physical downlink control channel (PDCCH) and a physical downlink shared channel (PDSCH).

[0072] In the user plane and control plane, Layer 2 (L2) of AS 304 may be responsible for the link above the physical layer 306 between the wireless device 320 and the base station 350. In various embodiments, Layer 2 may include a medium access control (MAC) sublayer 308, a radio link control (RLC) sublayer 310, and a packet data convergence protocol (PDCP) 312 sublayer, each of which forms a logical connection that terminates at the base station 350.

[0073] In the control plane, Layer 3 (L3) of AS 304 may include a radio resource control (RRC) sublayer 3. Although not shown, software architecture 300 may include additional Layer 3 sublayers and various upper layers above Layer 3. In various embodiments, RRC sublayer 313 may provide functionality including broadcasting system information, paging, and establishing and releasing RRC signaling connections between wireless devices 320 and base stations 350.

[0074] In various embodiments, the PDCP sublayer 312 may provide uplink functions including multiplexing between different radio bearers and logical channels, sequence number addition, handover data processing, integrity protection, ciphering, and header compression. In the downlink, the PDCP sublayer 312 may provide functions including in-sequence delivery of data packets, duplicate data packet detection, integrity verification, deciphering, and header decompression.

[0075] In the uplink, the RLC sublayer 310 may provide segmentation and concatenation of upper layer data packets, retransmission of lost data packets, and automatic repeat request (ARQ). In the downlink, the RLC sublayer 310 functions may include reordering of data packets to compensate for out-of-order reception, reassembly of upper layer data packets, and ARQ.

[0076] In the uplink, the MAC sublayer 308 can provide functions including multiplexing between logical channels and transport channels, random access procedures, logical channel priority, and hybrid ARQ (HARQ) operations. In the downlink, MAC layer functions may include channel mapping within the cell, demultiplexing, discontinuous reception (DRX), and HARQ operations.

[0077] While the software architecture 300 may provide functionality for sending data over a physical medium, the software architecture 300 may also include at least one host layer 314 to provide data transmission services to various applications in the mobile device 320. In some embodiments, the application-specific functionality provided by the at least one host layer 314 may provide an interface between the software architecture and the processor.

[0078] In other embodiments, the software architecture 300 may include one or more higher logical layers (e.g., transport, session, presentation, application, etc.) that provide host layer functionality. For example, in some embodiments, the software architecture 300 may include a network layer (e.g., an Internet Protocol (IP) layer) in which a logical connection terminates at a packet data network (PDN) gateway (PGW). In some embodiments, the software architecture 300 may include an application layer in which a logical connection terminates at another device (e.g., an end-user device, a server, etc.). In some embodiments, the software architecture 300 may also include a hardware interface 316 between the physical layer 306 and communication hardware (e.g., one or more radio frequency (RF) transceivers) in the AS 304.

[0079] Figure 4A and 4B is a component block diagram illustrating a system 400 configured for managing information transmission for wireless communications according to various embodiments. Figure 1-4B , the system 400 may include a base station 402 (e.g., 120a-120e, 200, 320) and a wireless device 404 (e.g., 120a-120e, 200, 320). The base station 402 and the wireless device 404 may communicate with each other via a wireless communication network 424 (aspects of which are described in detail in the accompanying drawings). Figure 1 ) for communication.

[0080] The base station 402 and / or wireless device 404 may include one or more processors 428, 432 (e.g., 210, 212, 214, 216, 218, 252, 260) coupled to electronic storage 426, 430 and a wireless transceiver 266. The wireless transceiver 266 may be configured to receive messages to be sent in uplink transmissions from the processors 428, 432 and transmit such messages via an antenna (not shown) to the wireless communication network 424 for relaying to the base station 402 and / or wireless device 404. Similarly, the wireless transceiver 266 may be configured to receive messages from the base station 402 and / or wireless device 404 in downlink transmissions from the wireless communication network 424 and pass the messages to the one or more processors 428, 432 (e.g., via a modem (e.g., 252) that demodulates the messages).

[0081] Processors 428, 432 may be configured by machine-readable instructions 406, 434. Machine-readable instructions 406 may include one or more instruction modules. Instruction modules may include computer program modules. Instruction modules may include one or more of the following: neural network training modules 410, 436, PRT neural network module 412, enhancement neural network module 416, receiver neural network module 438, wireless communication modules 414, 440, or other instruction modules.

[0082] The neural network training modules 410, 436 may be configured to train the PRT neural network and / or the receiver neural network on a data set of data tones and for accurate reconstruction of the data tones to within an error threshold.

[0083] The PRT neural network module 412 can be configured to apply a PRT neural network to the data tones to generate a PRT for combining with the data tones to reduce the PAPR of the transmission waveform. The PRT neural network can be trained to generate the PRT so that the trained receiver neural network can accurately generate a reconstruction of the data tones from a transmission waveform consisting of a combination of the data tones and the PRT.

[0084] The augmented neural network module 416 can be configured to apply an augmented neural network to augment the output of the PRT neural network. The augmented neural network can correct the output generated by the PRT neural network. The augmented neural network can compensate for the difference between the assumptions used in designing the PRT neural network (which can implement traditional PRT calculation algorithms) and the realities of circuits and environments that the PRT neural network was not designed to fully account for.

[0085] The receiver neural network module 438 can be configured to apply a receiver neural network to the transmission waveform to accurately generate a reconstruction of the data tones. The receiver neural network can be trained so that the PRT generated from the trained PRT neural network provides information to the trained receiver neural network to assist in accurately generating a reconstruction of the data tones from the transmission waveform composed of the combination of the data tones and the PRT.

[0086] The wireless communication modules 414 , 440 may be configured to transmit indicators of the PRT neural network and / or receiver neural network, weights of the PRT neural network and / or receiver neural network, and / or transmission waveforms between the wireless device 404 and the base station 402 .

[0087] In some embodiments, base station 402 and wireless device 404 can be operatively linked via one or more electronic communication links. For example, such electronic communication links can be established at least in part via a network such as the Internet and / or other networks. However, this example is not intended to be limiting, and the scope of the present disclosure includes embodiments in which base station 402 and wireless device 404 can be operatively linked via some other communication medium.

[0088] Electronic storage 426, 430 may include non-transitory storage media that electronically stores information. The electronic storage media of electronic storage 426, 430 may include one or both of the following: system storage provided integrally with base station 402 or wireless device 404 (i.e., substantially non-removably); and / or removable storage that is removably connected to base station 402 or wireless device 404 via, for example, a port (e.g., a universal serial bus (USB) port, a firewire port, etc.) or a drive (e.g., a disk drive, etc.). Electronic storage 426, 430 may include one or more of the following: optically readable storage media (e.g., optical disks, etc.), magnetically readable storage media (e.g., magnetic tape, a magnetic hard drive, a floppy disk drive, etc.), charge-based storage media (e.g., EEPROM, RAM, etc.), solid-state storage media (e.g., flash drives, etc.), and / or other electronically readable storage media. Electronic storage 426, 430 may include one or more virtual storage resources (e.g., cloud storage, a virtual private network, and / or other virtual storage resources). Electronic storage 426, 430 may store software algorithms, information determined by processors 428, 432, information received from base station 402 or wireless device 404, respectively, or other information that enables base station 402 or wireless device 404 to function as described herein.

[0089] Processors 428, 432 may be configured to provide information processing capabilities in base station 402. Thus, processors 428, 432 may include one or more of the following: a digital processor, an analog processor, a digital circuit designed to process information, an analog circuit designed to process information, a state machine, and / or other mechanisms for electronically processing information. Although processors 428, 432 are shown as a single entity, this is for illustrative purposes only. In some embodiments, processors 428, 432 may include multiple processing units and / or processor cores. The processing units may be physically located within the same device, or processors 428, 432 may represent processing functionality of multiple devices operating in coordination. Processors 428, 432 may be configured to execute modules 410-416 and modules 436-440 and / or other modules via: software; hardware; firmware; some combination of software, hardware, and / or firmware; and / or other mechanisms for configuring processing capabilities on processors 428, 432. As used herein, the term "module" may refer to any component or collection of components that performs the functionality attributed to the module. This may include one or more physical processors, processor-readable instructions, circuits, hardware, storage media, or any other components during the execution of processor-readable instructions.

[0090] The description of the functionality provided by the various modules 410-416 and modules 436-440 described herein is for illustrative purposes and is not intended to be limiting, as any of the modules 410-416 and modules 436-440 may provide more or less functionality than described. For example, one or more of the modules 410-416 and modules 436-440 may be eliminated, and some or all of their functionality may be provided by other modules 410-416 and modules 436-440. As another example, the processors 428, 432 may be configured to execute one or more additional modules that may perform some or all of the functionality attributed below to one of the modules 410-416 and modules 436-440.

[0091] Figure 5A A functional block diagram is shown including an example encoder circuit 500 configured to implement machine learning and / or machine learning-based encoding for combining a time-domain data signal derived from frequency-domain data tones and a time-domain PRT into a transmission waveform, according to various embodiments. Figure 1-5A, a transmitter (e.g., wireless devices 120a-120e, 200, 320, 404) may include an encoder 500. The encoder 500 may be implemented in hardware, software executed on a processor, and / or a combination of hardware and software running on a processor. The encoder 500 may be a standalone component of the transmitter, a component of a SoC (e.g., SoC 202, 204), and / or a component hardware and / or software component of a processor (e.g., processors 210, 212, 214, 216, 218, 252, 260, 428). The encoder 500 may include any number and combination of PRT neural networks 502a, 502b, enhanced neural networks ( Figure 5A The encoder 500 may include a plurality of signal summers 504a, 504b, 520, and an inverse fast Fourier transform component 506. The encoder 500 may include a plurality of signal summers 504a, 504b, 520, and an inverse fast Fourier transform component 506. The encoder 500 may use machine learning trained PRT neural networks 502a, 502b and a boosted neural network 508 configured to map frequency domain data tones and PRTs for transmission from a particular transmitter and / or to a particular receiver.

[0092] Figure 5B A functional block diagram is shown including an example enhanced neural network 508 configured to implement neural enhancement for machine learning and / or machine learning-based encoding for combining a time domain data signal derived from a frequency domain data tone and a time domain PRT into a transmission waveform, in accordance with various embodiments. Figure 1-5B , a transmitter (e.g., wireless devices 120a-120e, 200, 320, 404) may include an enhanced neural network 508. The enhanced neural network 508 may be implemented in hardware, software executed on a processor, and / or a combination of hardware and software running on a processor. The enhanced neural network 508 may be a standalone component of the transmitter, a component of a SoC (e.g., SoC 202, 204), and / or a hardware and / or software component of a processor (e.g., processors 210, 212, 214, 216, 218, 252, 260, 428). The enhanced neural network 508 may include a feature map generator 510 and neural network layers 512a, 512b, 514a, 514b, 514c.

[0093] The enhanced neural network 508 may be a machine-learning trained neural network configured to enhance the output of the previous PRT neural networks 502a, 502b. In some embodiments, the enhanced neural network 508 may correct the output generated by a conventional PRT calculation algorithm, or the output generated by the previous PRT neural networks 502a, 502b configured to implement an expansion of the conventional PRT calculation algorithm. The performance of the previous PRT neural networks 502a, 502b may be limited by the conventional PRT calculation algorithm implemented by the PRT neural networks 502a, 502b. Enhancing the output of the previous PRT neural networks 502a, 502b may enhance the performance of the PRT neural networks 502a, 502b. This may be achieved by configuring the enhanced neural network 508 to modify the output of the conventional PRT calculation algorithm generated by the PRT neural networks 502a, 502b to suppress any power peaks remaining in the transmission waveform output from the neural networks 502a, 502b. Augmenting the outputs of the previous PRT neural networks 502a, 502b by the augmenting neural network 508 may increase the power and / or mitigate nonlinearities in the time-domain transmission waveform generated by the encoder 500.

[0094] Specifically, the enhanced neural network 508 can compensate for the discrepancies between the assumptions used when designing the conventional PRT calculation algorithm and the realities of the circuit and environment that the conventional PRT calculation algorithm was not designed to fully account for. For example, in circuits and / or environments where the conventional PRT calculation algorithm fails to perform well, the neural network enhancements implemented by the enhanced neural network 508 can implement functions that the conventional PRT calculation algorithm fails to implement, thereby making the entire system more robust by further reducing the PAPR. As another example, the enhanced neural network 508 can provide functions not captured by the PRT neural networks 502a, 502b, such as compensating for missing primary components in the data path, enhancing the embedding of distortion information in the time-domain transmission waveform to improve decoder gain, compensating for any uncompensated non-idealities in the RF chain that the engineer failed to capture in the design, and the like.

[0095] Figure 6A An example decoder 600 is shown that is configured to implement machine learning and / or machine learning-based decoding to extract PRT from a received transmission waveform and use information from this process to improve the performance of a receiver in estimating the original data pitch, in accordance with various embodiments. Figure 1-6A, a receiver (e.g., base stations 110a-110d, 350, 402) may include a decoder 600. The decoder 600 may be implemented in hardware, software executed on a processor, and / or a combination of hardware and software running on a processor. The decoder 600 may be a standalone component of the receiver, a component of a SoC (e.g., SoC 202, 204), and / or a component hardware and / or software component of a processor (such as processors 210, 212, 214, 216, 218, 252, 256, 260, 432). The decoder 600 may include a fast Fourier transform component 602 and a receiver neural network 604. The decoder 600 may use a neural network trained by machine learning that is configured to demodulate a transmission waveform received from a particular transmitter and / or received by a particular receiver, the transmission waveform comprising a time domain data signal derived from frequency domain data tones and a time domain PRT.

[0096] Figure 6B An example receiver neural network 604 is shown that is configured to implement machine learning and / or machine learning-based decoding to extract PRT from a received transmission waveform and use information from this process to improve the receiver's performance in estimating the original data pitch, in accordance with various embodiments. Figure 1-6B , a receiver (e.g., base stations 110a-110d, 350, 402) may include a receiver neural network 604. The receiver neural network 604 may be implemented in hardware, software executed on a processor, and / or a combination of hardware and software running on a processor. The receiver neural network 604 may be a standalone component of the receiver, a component of a SoC (e.g., SoC 202, 204), and / or a component hardware and / or software component of a processor (such as processors 210, 212, 214, 216, 218, 252, 256, 260, 432). The receiver neural network 604 may include a channel equalizer component 610, a time domain PRT extractor component 612, neural network layers 614a, 614b, 616a, 616b, and a signal adder 618. Receiver neural network 604 may be a neural network trained via machine learning configured to demodulate a transmission waveform received from a particular transmitter and / or received by a particular receiver, the transmission waveform comprising a time-domain data signal and a time-domain PRT derived from frequency-domain data tones.

[0097] Reference together Figure 5A 、 5B, 6A, and 6B, machine learning can be used to train the set of PRT neural networks 502a, 502b and the enhanced neural network 508, as well as the receiver neural network 604. The machine learning method can be implemented separately at the transmitter and / or the receiver, during which the transmitter and / or the receiver can implement the set of PRT neural networks 502a, 502b, the enhanced neural network 508, and the receiver neural network 604. The set of PRT neural networks 502a, 502b and the enhanced neural network 508 can be trained to receive as input a time-domain data signal derived from the frequency-domain data tones and output a transmission waveform based on a combination of the time-domain PRT and the time-domain data signal and / or a combined time-domain data signal generated from a previous combination in the signal summers 504a, 504b. The time-domain data signal can be derived from the frequency-domain data tones by transforming the frequency-domain data tones into the time-domain data signal via the inverse fast Fourier transform component 506. The result of the successive combinations of the time-domain PRT with the time-domain data signal and the combined time-domain data signal is a transmit waveform suitable for amplification for transmission with reduced peaking to reduce PAPR. On the receiver side, a receiver neural network 604 can be trained to demodulate the transmit waveform, which is a combination of the frequency-domain data tones represented in the time domain by the time-domain data signal and the time-domain PRT, to generate a reconstruction of the frequency-domain data tones based on the transmit waveform input.

[0098] In a non-limiting example, the set of PRT neural networks 502a, 502b, the boosted neural network 508, and the pair of receiver neural networks 604 can be implemented as autoencoders using unsupervised machine learning. The set of PRT neural networks 502a, 502b, the boosted neural network 508, and the pair of receiver neural networks 604 can be trained so that the set of PRT neural networks 502a, 502b and / or the boosted neural network 508 generate a PRT to be combined with the time-domain data signal, such that the receiver neural network 604 can accurately generate a reconstruction of the time-domain data signal, such as within an error threshold. In some embodiments, the error can be determined by comparing a reconstruction of the frequency-domain data tones derived from the reconstruction of the time-domain data signal with the frequency-domain data tones. In some embodiments, the error in the reconstruction of the frequency-domain data tones can be calculated as the mean squared error between the frequency-domain data tones and the reconstruction of the frequency-domain data tones. In some embodiments, the error can be determined by outputting a distortion function based on energy outside of an allocated frequency band. In some embodiments, the total error can be obtained by a weighted average of the error (error1) from the comparison of the reconstruction of the frequency domain data tone with the frequency domain data tone and the error (error2) from the output of the distortion function. For example, for a certain "α" that can be found during hyperparameter tuning, the total error can be = error1 + α * error2.

[0099] In some embodiments, the set of PRT neural networks 502a, 502b, the boosting neural network 508, and the receiver neural network 604 may be trained for a specific transmitter, such as by using the hardware configuration of the transmitter.

[0100] In some embodiments, the set of PRT neural networks 502a, 502b, the enhanced neural network 508, and the receiver neural network 604 can be trained for a specific receiver, such as by using the receiver's hardware configuration. For example, the set of PRT neural networks 502a, 502b, the enhanced neural network 508, and the receiver neural network 604 pair can be trained using a specific transmitter and / or a transmitter similar to the specific transmitter (such as a transmitter using the same hardware and / or software configuration, such as a reference transmitter). As a further example, the set of PRT neural networks 502a, 502b, the enhanced neural network 508, and the receiver neural network 604 pair can be trained using a specific receiver and / or a receiver similar to the specific receiver (such as a receiver using the same hardware and / or software configuration, such as a reference receiver).

[0101] The transmitter and the receiver may share the configuration of the set of PRT neural networks 502a, 502b, the enhanced neural network 508, and / or the receiver neural network 604. The transmitter may share the configuration of the receiver neural network 604 by wirelessly transmitting the configuration of the receiver neural network 604 to the receiver. For example, the transmitter may share the configuration of the receiver neural network 604 with the receiver for the receiver to implement, which may be the receiver neural network 604 trained in conjunction with the set of PRT neural networks 502a, 502b and the enhanced neural network 508 implemented by the transmitter.

[0102] In some embodiments, a receiver may be pre-configured with multiple receiver neural networks 604. The configuration of the shared receiver neural network 604 may include an indicator of the configuration of the shared receiver neural network 604. The receiver may use the indicator to select the configuration of the receiver neural network 604 from among the multiple receiver neural networks 604.

[0103] In some embodiments, configuration of the shared receiver neural network 604 may include sharing weights resulting from training the set of PRT neural networks 502a, 502b and the boosted neural network 508 and / or the receiver neural network 604. Configuration of the shared receiver neural network 604 may include sending an indicator and / or weights to the receiver. In some embodiments, the weights may be indicators.

[0104] The receiver can share the configuration of the set of PRT neural networks 502a, 502b and the enhanced neural network 508 by wirelessly transmitting the configuration of the set of PRT neural networks 502a, 502b and the enhanced neural network 508 to the transmitter. For example, the receiver can share the configuration of the set of PRT neural networks 502a, 502b and the enhanced neural network 508 with the transmitter for implementation by the transmitter, which configuration can be the set of PRT neural networks 502a, 502b and the enhanced neural network 508 trained in conjunction with the receiver neural network 604 implemented by the receiver. In some embodiments, the transmitter can be pre-configured with multiple sets of PRT neural networks 502a, 502b and multiple enhanced neural networks 508. Sharing the configuration of the set of PRT neural networks 502a, 502b and the enhanced neural network 508 can include an indicator of the shared configuration of the set of PRT neural networks 502b, 502a and the enhanced neural network 508. The transmitter can use the indicator to select a configuration of a set of PRT neural networks 502a, 502b and a boosted neural network 508 from a plurality of sets of PRT neural networks 502a, 502b and a boosted neural network 508. In some embodiments, sharing the configuration of the set of PRT neural networks 502a, 502b and the boosted neural network 508 can include sharing weights resulting from training the set of PRT neural networks 502a, 502b and the boosted neural network 508 and / or the receiver neural network 604. Sharing the configuration of the set of PRT neural networks 502a, 502b and the boosted neural network 508 can include sending the indicator and / or weights to the transmitter. In some embodiments, the weights can be indicators.

[0105] The transmitter can use the set of PRT neural networks 502a, 502b and the boosted neural network 508 to generate a time-domain PRT to reduce the PAPR within the transmitter, and the receiver can then use the corresponding trained receiver neural network 604 to demodulate (i.e., decode) the transmitted waveform received from the transmitter. In various embodiments, the transmitter can use the set of trained PRT neural networks 502a, 502b and the boosted neural network 508 to generate a time-domain PRT as the frequency-domain data tones for transmission are received (i.e., generate the PRT "on the fly"). By using the receiver neural network 604 trained in conjunction with (or using the output from) the transmitter's set of PRT neural networks 502a, 502b and the boosted neural network 508, the receiver can more accurately reconstruct the frequency-domain data tones from the received waveform (compared to what can be achieved using conventional demodulation circuitry). To achieve this benefit, some embodiments include operations for coordinating between the transmitter and the receiver so that the set of trained PRT neural networks 502a, 502b and the enhanced neural network 508 used in the transmitter corresponds to the trained receiver neural network 604 used in the receiver, and vice versa. In some embodiments, the transmitter may use a set of PRT neural networks 502a, 502b and the enhanced neural network 508 selected in response to receiving an indicator of the configuration of the set of PRT neural networks 502a, 502b and the enhanced neural network 508. In some embodiments, the transmitter may use a set of PRT neural networks 502a, 502b and the enhanced neural network 508 that is pre-configured on the transmitter.

[0106] The inverse fast Fourier transform component 506 of the transmitter can receive frequency-domain data tones and convert the frequency-domain data tones into a time-domain data signal. A first PRT neural network 502a can receive the time-domain data signal and generate a time-domain PRT for the time-domain data signal. A first signal adder 504a can combine the time-domain data signal with the PRT generated by the first PRT neural network 502a to generate a first combined time-domain signal. A subsequent PRT neural network, such as the second PRT neural network 502b, can receive the first combined time-domain signal and generate a time-domain PRT for the first combined time-domain signal. A second signal adder 504b can combine the first combined time-domain signal with the PRT generated by the second PRT neural network 502b to generate a second combined time-domain signal. In some embodiments, a subsequent PRT neural network (not shown) can receive the second combined time-domain signal and generate a time-domain PRT for the second combined time-domain signal. Successive signal adders (not shown) may combine the second combined time-domain signal with the PRT generated by the successive PRT neural networks to generate a third combined time-domain signal. In some embodiments, further successive PRT neural networks may continue to generate the pattern of the PRT for the previous combined time-domain signal, and further successive signal adders may combine the PRT with the previous combined time-domain signal to generate successive combined time-domain signals.

[0107] In some embodiments, the PRT neural networks 502a, 502b may implement conventional PRT calculation algorithms. For example, the PRT neural networks 502a, 502b may generate outputs according to the following equations:

[0108]

[0109] Wherein, “x” is the vector of time domain samples input to the PRT neural network 502a, 502b; “x(l)” is the lth element of the vector “x”; “N FFT" is the size of the inverse Fourier transform for each OFDM symbol; "p" is a kernel vector corresponding to the PRT tone position; "A" and "μ" are values ​​learned during training, and their values ​​may be specific to the PRT neural networks 502a, 502b. Each PRT neural network 502a, 502b may correspond to an iteration of a conventional PRT calculation algorithm. However, unlike conventional PRT calculation algorithms (where the values ​​of "A" and "μ" remain the same for each iteration of the algorithm), PRT neural networks 502a, 502b based on conventional PRT calculation algorithms will learn values ​​of "A" and "μ" that may change from iteration to iteration (e.g., from PRT neural network 502a to PRT neural network 502b) based on the results of training. The kernel vector "p" may constrain the PRT neural networks 502a, 502b to generate a correction to the previous PRT value at each iteration, thereby generating a new PRT value.

[0110] The enhancement neural network 508 may be configured to receive and enhance the final combined time domain signal from the series of PRT neural networks 502a, 502b and signal adders 504a, 504b. The feature map generator 510 may receive the final combined time domain signal (x n ) and generates a feature map 516 of the final combined time domain signal (such as two real feature maps or one complex feature map) and a feature map 518 of the function of the final combined time domain signal (such as function x n |x n | 2 The feature maps 516, 518 may be input to a first neural network layer 512a of the enhanced neural network 508, such as a one-dimensional convolutional neural network layer with a 3-channel kernel. The output of the first neural network layer 512a may be input to a second neural network layer 514a, such as a residual neural network (ResNet) block with a 3-channel kernel. The output of the second neural network layer 514a may be input to a third neural network layer 514b, such as a ResNet block with a 3-channel kernel. The output of the third neural network layer 514b may be input to a fourth neural network layer 514c, such as a ResNet block with a 3-channel kernel. The output of the fourth neural network layer 514c may be input to a fifth neural network layer 512b, such as a one-dimensional convolutional neural network layer with a 1-channel kernel. The signal adder 520 may combine the output of the fifth neural network layer 512b with the final combined time domain signal to generate a time domain transmission waveform to be sent to the receiver. In some embodiments, the neural network layers 512a, 512b, 514a, 514b, 514c of the boosted neural network 508 may be configured to implement circular convolutions.

[0111] The receiver can receive a transmission waveform in the time domain and can use a receiver neural network 604 to demodulate the transmission waveform to generate a reconstruction of the time domain data signal. The receiver neural network 604 can receive a transmission waveform in the time domain and demodulate the transmission waveform to generate a reconstruction of the time domain data signal. In some embodiments, the receiver can use a receiver neural network 604 that is pre-configured on the receiver. In some embodiments, the receiver can use a receiver neural network 604 that is selected in response to receiving an indicator of the configuration of the receiver neural network 604. The fast Fourier transform component 602 of the receiver can convert the reconstruction of the time domain data signal from a reconstruction of the time domain data signal to a reconstruction of the frequency domain data tones.

[0112] In order to demodulate the transmission waveform by the receiver neural network 604 to generate a reconstruction of the time domain data signal, the channel equalizer component 610 equalizes the transmission waveform. The channel equalizer component 610 can apply any known equalization process. Equalization of the transmission waveform can eliminate the effects of the propagation channel on the frequency response of the transmission waveform. The channel equalizer component 610 can generate various signals from the equalized transmission waveform, including: equalized time domain signals 624, 628 (y n ); a characteristic diagram of the equalized time domain signal 620, such as two real characteristic diagrams or a complex characteristic diagram; and a characteristic diagram of a function of the equalized time domain signal 622, such as a function y n |y n | 2 The time domain PRT extractor component 612 can receive the equalized time domain signal 624 and extract the time domain PRT (p n ), and generates a feature map (such as two real feature maps or one complex feature map) of a time-domain PRT 626 from the equalized time-domain signal 624. A first neural network layer 614a (e.g., a one-dimensional convolutional neural network layer with a 3-channel kernel) can receive as input the feature map of the equalized time-domain signal 620, the feature map of the function of the equalized time-domain signal 622, and the time-domain PRT 626. The output of the first neural network layer 614a can be input to a second neural network layer 616a (e.g., a residual neural network (ResNet) block with a 3-channel kernel). The output of the second neural network layer 616a can be input to a third neural network layer 616b (e.g., a ResNet block with a 3-channel kernel). The output of the third neural network layer 616b can be input to a fourth neural network layer 614b (e.g., a one-dimensional convolutional neural network layer with a 1-channel kernel). The signal summer 618 may combine the output of the fourth neural network layer 614b with the equalized time domain signal 628 to generate a reconstruction of the time domain data signal for input to the fast Fourier transform component 602 to generate a reconstruction of the frequency domain data tones.

[0113] For the set of PRT neural networks 502a, 502b, the boosting neural network 508, and the receiver neural network 604 that are trained together, the means for generating a time-domain PRT by the set of PRT neural networks 502a, 502b and the boosting neural network 508 for combining with the time-domain data signal can be known to the receiver implementing the receiver neural network 604. As such, the time-domain PRT can contain information for demodulating a transmitted waveform that the trained receiver neural network 604 can use to generate a reconstructed frequency-domain data tone. The information contained in the time-domain PRT can enable the receiver to use the trained receiver neural network 604 to improve demodulation performance, such as improving the accuracy of reconstruction of the frequency-domain data tone compared to transmitting a transmitted waveform with a PRT generated by a conventional signal processing algorithm and demodulating using conventional demodulation circuitry.

[0114] Figure 7 An example of a system configured to implement machine learning-based encoding and decoding for nonlinearity mitigation using PRT according to various embodiments is shown. Figure 1-7 , a transmitter 700 (e.g., wireless devices 120a-120e, 200, 320, 404) may indicate to a receiver 702 (e.g., base stations 110a-110d, 350, 402) a receiver neural network (e.g., receiver neural network 604) to use to decode a transmitted waveform to generate a reconstruction of a time-domain data signal for use in generating a reconstruction of frequency-domain data tones.

[0115] Transmitter 700 and receiver 702 may establish an uplink 704. Transmitter 700 may send a receiver neural network indicator 706 to receiver 702. The receiver neural network indicator may be configured to indicate to receiver 702 a receiver neural network to be selected from a plurality of receiver neural networks to decode a transmission waveform from transmitter 700. In some embodiments, the receiver neural network indicator may be a reference configured to indicate to receiver 702 the receiver neural network to be selected, such as a flag bit in a signal, a value content in a signal, a quality of a signal, etc. In some embodiments, the receiver neural network indicator may be an indicator of the receiver neural network to be selected. In some embodiments, the receiver neural network indicator may be an indicator of the set of PRT neural networks (e.g., PRT neural networks 502a, 502b) and the enhanced neural network (e.g., enhanced neural network 508) that transmitter 700 is configured to utilize to generate a transmission waveform, and receiver 702 may determine the receiver neural network to be selected based on the indicator. In some embodiments, the receiver neural network indicator may be weights for use with the receiver neural network, and the receiver 702 may determine the receiver neural network to select based on these weights. In some embodiments, the receiver 702 may use the receiver neural network indicator as a value for a hash function, a lookup table, a data structure location, etc., for selecting a receiver neural network.

[0116] In some embodiments, the transmitter can be configured to select from a set of neural network implementations based on whether the transmitter is capable of implementing an enhanced neural network and / or a receiver neural network. Based on whether the transmitter is capable of implementing the enhanced neural network and / or the receiver neural network, the transmitter can select a neural network implementation and send weights corresponding to the neural network selected for implementation to the receiver in uplink control information. Which neural network implementation the transmitter selects can depend on the transmitter determining which neural network implementation is optimal for the modulation and coding scheme. In some embodiments, in response to determining that the transmitter will implement the enhanced neural network and the receiver neural network, the neural network implementation can include using a set of PRT neural networks and using any combination of the enhanced neural network and / or the receiver neural network. In some embodiments, in response to determining that the transmitter will implement the enhanced neural network and not implement the receiver neural network, the neural network implementation can include using a set of PRT neural networks and using or not using the enhanced neural network.

[0117] The transmitter 700 may send weights 708 for a receiver neural network to the receiver 702. The weights may be weights for the receiver 702 to use in implementing a selected receiver neural network for demodulating the transmitted waveform to generate a reconstruction of the time-domain data signal for use in generating a reconstruction of the frequency-domain data tones. In some embodiments, the weights may be receiver neural network indicators, and sending the weights 708 for the receiver neural network may be combined with sending the receiver neural network indicator 706.

[0118] Transmitter 700 can encode a time-domain data signal derived from frequency-domain data tones using a time-domain PRT using a set of PRT neural networks and a boosting neural network 710. An inverse fast Fourier transform component (e.g., inverse fast Fourier transform component 506) can receive the frequency-domain data tones and convert them into a time-domain data signal. The set of PRT neural networks and the boosting neural network of transmitter 700 used to encode the time-domain data signal can be trained in conjunction with a receiver neural network. The set of PRT neural networks and the boosting neural network can receive the time-domain data signal and generate a PRT based on the training of the set of PRT neural networks, the boosting neural network, and the receiver neural network. This training can result in the receiver neural network accurately generating a reconstruction of the frequency-domain data signal. A signal summer (e.g., signal summers 504a, 504b) can receive and combine the time-domain data signal and the PRT. A signal summer (e.g., signal summer 520) can receive and combine the final combined time-domain signal and the output of the boosting neural network to generate a transmission waveform. The transmitter 700 may send a transmission waveform 712 in the time domain to the receiver 702 .

[0119] Receiver 702 can receive the transmission waveform and decode the transmission waveform 714 using a receiver neural network and weights. The receiver neural network used to demodulate the transmission waveform can be a receiver neural network selected based on an indicator received from transmitter 700. The weights that receiver 702 can use with the receiver neural network can be weights received from transmitter 700. The receiver neural network can receive the transmission waveform in the time domain and demodulate the transmission waveform using the received weights. The demodulated transmission waveform can generate a reconstruction of the time domain data signal. A fast Fourier transform component (e.g., fast Fourier transform component 602) can receive the reconstruction of the data signal in the time domain and convert the reconstruction of the time domain data signal into reconstructed frequency domain data tones.

[0120] Figure 8 An example of a system configured to implement machine learning-based encoding and decoding for nonlinearity mitigation using PRT according to various embodiments is shown. Figure 1-8, a receiver 702 (e.g., base stations 110a-110d, 350, 402) may indicate to a transmitter 700 (e.g., wireless devices 120a-120e, 200, 320, 404) a set of PRT neural networks (e.g., PRT neural networks 502a, 502b) and a boosting neural network (e.g., boosting neural network 508) to use to encode a time-domain data signal derived from frequency-domain data tones using time-domain PRT to generate a transmission waveform.

[0121] Transmitter 700 and receiver 702 may establish an uplink 704. Receiver 702 may send a PRT neural network indicator 800 to transmitter 700. The PRT neural network indicator may be configured to indicate to transmitter 700 a set of PRT neural networks and an enhanced neural network to be selected by transmitter 700 from a plurality of PRT neural networks to encode a time-domain data signal derived from frequency-domain data tones using a time-domain PRT. In some embodiments, the PRT neural network indicator may be a reference configured to indicate to transmitter 700 the set of PRT neural networks and the enhanced neural network to be selected, such as a flag bit in a signal, a value content in a signal, a quality of a signal, etc. In some embodiments, the PRT neural network indicator may be an indicator of the set of PRT neural networks and the enhanced neural network to be selected. In some embodiments, the PRT neural network indicator may be an indicator of the receiver neural network (e.g., receiver neural network 604) that receiver 702 is configured to utilize to decode a transmission waveform, from which transmitter 700 may determine the PRT neural network to be selected. In some embodiments, the PRT neural network indicator may be weights for use with the PRT neural network set and the enhanced neural network, and the transmitter 700 may determine the PRT neural network set and the enhanced neural network to select based on these weights. In some embodiments, the transmitter 700 may use the PRT neural network indicator as a value for a hash function, a lookup table, a data structure location, etc. for selecting the PRT neural network set and the enhanced neural network.

[0122] The receiver 702 may transmit weights 802 for the PRT neural network set and the enhanced neural network to the transmitter 700. The weights may be weights for the transmitter 700 to use when implementing the selected PRT neural network set and the enhanced neural network for generating the PRT to generate the transmission waveform. In some embodiments, the weights may be PRT neural network indicators, and transmitting the weights 802 for the PRT neural network set and the enhanced neural network may be combined with transmitting the PRT neural network indicator 800.

[0123] The transmitter 700 can use a set of PRT neural networks, a boosted neural network, and weights to encode a time-domain data signal derived from frequency-domain data tones using a time-domain PRT 804. An inverse fast Fourier transform component (e.g., inverse fast Fourier transform component 506) can receive the data tones in the frequency domain and convert the data tones into a time-domain data signal. The set of PRT neural networks and the boosted neural network that the transmitter 700 can use to encode the time-domain data signal derived from the frequency-domain data tones using the time-domain PRT can be a set of PRT neural networks and a boosted neural network selected based on an indicator received from the receiver 702. The weights that the transmitter 700 can use with the set of PRT neural networks and the boosted neural network to encode the time-domain data signal derived from the frequency-domain data tones using the time-domain PRT can be the weights received from the receiver 702. The PRT neural network set and the boosted neural network can receive a time domain data signal derived from the frequency domain data tones and use the received weights to generate a time domain PRT based on training the PRR neural network set, the boosted neural network, and the receiver neural network. This training can cause the receiver neural network to accurately generate a reconstruction of the time domain data signal, which can be used to reconstruct the frequency domain data tones. A signal adder (e.g., signal adders 504a, 504b) can receive and combine the time domain data signal and the time domain PRT. A signal adder (e.g., signal adder 520) can receive and combine the final combined time domain signal and the output of the boosted neural network to generate a transmission waveform. The transmitter 700 can send a transmission waveform 712 in the time domain to the receiver 702.

[0124] Receiver 702 can receive a transmission waveform and use a receiver neural network to decode the transmission waveform 806. The receiver neural network of receiver 702 can be used to decode the transmission waveform and can be a receiver neural network trained in conjunction with a set of PRT neural networks and a boosting neural network. The receiver neural network can receive the transmission waveform in the time domain and demodulate the transmission waveform to generate a reconstruction of the time domain data signal. As part of demodulating the received transmission waveform, the receiver neural network extracts the time domain PRT to obtain the time domain data signal, but also uses information inherent in the manner in which the PRT is generated by the set of PRT neural networks and the boosting neural network to improve the accuracy of the reconstructed time domain data signal. A fast Fourier transform component (e.g., fast Fourier transform component 602) can receive the reconstructed time domain data signal and convert the reconstructed time domain signal into reconstructed frequency domain data tones.

[0125] Figure 9 Methods for transmitting machine learning-based data for non-linearity mitigation using PRT according to various embodiments are shown. Figure 1-9, method 900 may be implemented in a computing device (e.g., base stations 110a-110d, 350, 402, wireless devices 120a-120e, 200, 320, 404, transmitter 700, receiver 702), in general-purpose hardware, in dedicated hardware, in software executed in a processor (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432), or in a combination of software-configured processors and dedicated hardware (such as processors (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432, encoder 500, decoder 600) and various memory / cache controllers) executing software within a peak reduction encoder and / or decoder system including other separate components. To encompass alternative configurations implemented in various embodiments, hardware implementing method 900 is referred to herein as a "wireless communication device."

[0126] In block 902, a wireless communication device may establish an uplink with another wireless communication device. Establishing the uplink may be initiated by the wireless communication device and may be established based on communication signals and data transmitted between the wireless communication devices. The uplink may be established via various known means, including means for establishing an uplink in a 5G NR network. In some embodiments, establishing the uplink in block 902 may occur between a transmitter and a receiver. In some embodiments, establishing the uplink in block 902 may occur between the wireless device and a base station.

[0127] In block 904, the wireless communication device may transmit the set of PRT neural networks, the enhanced neural network, and / or the receiver neural network. The receiver neural network may be a receiver neural network trained in conjunction with the set of PRT neural networks and the enhanced neural network for use by the wireless communication device. In some embodiments, in block 904, the transmitter and / or wireless device may transmit the receiver neural network.

[0128] In some embodiments, transmitting the PRT neural network set and the enhanced neural network in block 904 may be accomplished by transmitting an indicator of the PRT neural network set and the enhanced neural network, the indicator configured to indicate to the other wireless communication device which PRT neural network set (e.g., PRT neural network 502a, 502b) and enhanced neural network (e.g., enhanced neural network 508) to select from a plurality of PRT neural network sets pre-configured on the other wireless communication device. In some embodiments, transmitting the receiver neural network may be accomplished by transmitting an indicator of the receiver neural network (e.g., receiver neural network 604), the indicator configured to indicate to the other wireless communication device which receiver neural network to select from a plurality of receiver neural networks pre-configured on the other wireless communication device. In some embodiments, the indicator may be a reference configured to indicate to the other wireless communication device which PRT neural network set, enhanced neural network, and / or receiver neural network to select, such as a flag bit in a signal, the content of a value in a signal, the quality of a signal, etc. In some embodiments, the indicator may be an indicator of the PRT neural network set, enhanced neural network, and / or receiver neural network to select. In some embodiments, the indicator may be an indicator of the PRT neural network set and enhanced neural network that the transmitting wireless communication device is configured to utilize to generate the transmission waveform, and the receiving wireless communication device may determine the receiver neural network to select based on the indicator. In some embodiments, the indicator may be an indicator of the receiver neural network that the receiving wireless communication device is configured to utilize to decode the transmission waveform, and the transmitting wireless communication device may determine the PRT neural network set and enhanced neural network to use for generating the time-domain PRT based on the indicator. In some embodiments, the indicator may be a weight for use with the PRT neural network set, enhanced neural network, and / or receiver neural network, and another wireless communication device may determine the PRT neural network set, enhanced neural network, and / or receiver neural network to select based on these weights. In some embodiments, the other wireless communication device may use the indicator as a value for a hash function, a lookup table, a data structure location, etc. used to select the PRT neural network set, enhanced neural network, and / or receiver neural network.

[0129] In some embodiments, a wireless communication device may transmit a set of PRT neural networks and an enhanced neural network to another wireless communication device. The set of PRT neural networks and the enhanced neural network may be a set of PRT neural networks and an enhanced neural network trained in conjunction with a receiver neural network used by the wireless communication device. In some embodiments, in block 904, a receiver and / or a base station may transmit the set of PRT neural networks and the enhanced neural network. In some embodiments, a wireless communication device may transmit the receiver neural network to another wireless communication device.

[0130] In block 906, the wireless communication device may transmit the PRT neural network set, the enhanced neural network, and / or the receiver neural network weights. The wireless communication device may transmit the weights to another wireless communication device. The weights may be weights used by the transmitting wireless communication device when implementing the selected PRT neural network set and enhanced neural network for generating a time-domain PRT to generate a transmission waveform. In some embodiments, the weights may be PRT neural network indicators, and transmitting the weights for the PRT neural network set and enhanced neural network in block 906 may be combined with transmitting the PRT neural network indicators in block 904. In some embodiments, the weights may be weights used by the receiving wireless communication device when implementing the selected receiver neural network for demodulating the transmission waveform to generate a reconstruction of the frequency-domain data tones. In some embodiments, the weights may be receiver neural network indicators, and transmitting the weights for the receiver neural network in block 906 may be combined with transmitting the receiver neural network indicators in block 904. In some embodiments, the wireless communication device may transmit the PRT neural network to another wireless communication device. In some embodiments, the receiver and / or base station may transmit the PRT neural network weights in block 906. In some embodiments, the transmitter and / or wireless device may transmit the receiver neural network weights in block 906.

[0131] In some embodiments, method 900 may be implemented for each uplink established between a mobile wireless communication device and a base wireless communication device. In some embodiments, method 900 may be repeatedly implemented for each interval of a specific number of uplinks established between a wireless communication device and a base wireless communication device. In some embodiments, method 900 may be repeatedly implemented for uplinks established between a wireless communication device and a base wireless communication device after a specified time period.

[0132] Figure 10A An example of an end-to-end method 1000 for encoding frequency-domain data tones into waveforms sent by a transmitter and recovering the frequency-domain data tones in a receiver for machine learning-based encoding and decoding using nonlinearity mitigation of PRTs is shown in accordance with various embodiments. Figure 1-10A, method 1000 may be implemented in a computing device (e.g., base stations 110a-110d, 350, 402, wireless devices 120a-120e, 200, 320, 404, transmitter 700, receiver 702), in general-purpose hardware, in dedicated hardware, in software executed in a processor (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432), or in a combination of a software-configured processor and dedicated hardware (such as a processor (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432, encoder 500, decoder 600) and various memory / cache controllers) executing software within a peak reduction encoder and / or decoder system including other separate components. To encompass alternative configurations implemented in various embodiments, hardware implementing method 1000 is referred to herein as a "wireless communication device."

[0133] In block 1002, a transmitter, a transmitter wireless communication device, an encoder, and / or an inverse fast Fourier transform component of the transmitter wireless communication device (e.g., the inverse fast Fourier transform component 506) may receive frequency-domain data tones in block 1002. The data tones may be a data stream (such as a data packet) that has been mapped to a sequence of modulation symbols, such as quadrature amplitude modulation (QAM) symbols (e.g., 16QAM symbols).

[0134] In block 1004, a transmitter, transmitter wireless communication device, encoder, and / or inverse fast Fourier transform component of the transmitter wireless communication device may transform frequency domain data tones into time domain data signals for use in generating a transmit waveform for amplification and transmission.

[0135] In block 1006, the transmitter, transmitter wireless communication device, and / or encoder may generate a transmission waveform using a set of PRT neural networks (e.g., the set of PRR neural networks 502a, 502b) and a boosting neural network (e.g., boosting neural network 508) of the wireless communication device, as described herein with reference to Figure 121200 . In block 1004 , the transmitter wireless communication device may process the transformed time-domain data signal via the trained set of PRT neural networks and the enhanced neural network to generate a transmission waveform. As described herein, the set of PRT neural networks and the enhanced neural network in the transmitter may be trained based on the accuracy of reconstruction of frequency-domain data tones by a trained receiver neural network (e.g., receiver neural network 604). The set of PRT neural networks and the enhanced neural network may be trained to reduce the PAPR in a signal for transmission by the transmitting wireless communication device. By generating a PRT for the frequency-domain data signal via the set of PRT neural networks and the enhanced neural network, and including the PRT in the transmission waveform, the transmission waveform may include information inherent in the PRT that may assist the receiver neural network in more accurately recovering the time-domain data signal than would be possible using conventional demodulation circuitry.

[0136] In block 1008, a transmitter, a transmitter wireless communication device, a power amplifier of the transmitter wireless communication device (eg, resources 224 of the first SOC 202 including the power amplifier) ​​may amplify the transmission waveform. The amplification of the transmission waveform may amplify the transmission power of the transmission waveform.

[0137] In block 1010, a transmitter of a wireless communication device may transmit a transmission waveform to another wireless communication device. In some embodiments, the wireless communication device may transmit the transmission waveform to the wireless communication device via a 5G NR network.

[0138] In block 1012, an antenna of a receiver wireless communication device may receive a transmission waveform from a transmitting wireless communication device.

[0139] In optional block 1014, the receiving wireless communication device may select a receiver neural network (e.g., receiver neural network 604) for demodulating wireless signals received from the transmitter wireless communication device. The operations in block 1014 are optional because, in some embodiments, the receiver neural network may not change. Furthermore, the operations in optional block 1014 may be performed prior to receiving the transmission waveform, such as during a process for establishing a wireless communication link with the transmitter. In some embodiments, the receiver wireless communication device may have received and / or may receive the receiver neural network indicator as part of a transmission from the transmitter wireless communication device, as described herein in method 900( Figure 9) as described in block 904 of the preceding text. In some embodiments, the receiving wireless communication device may use the receiver neural network indicator as a value for a hash function, a lookup table, a data structure location, etc. for selecting a receiver neural network. In some embodiments, the receiving wireless communication device may be communicatively linked to multiple wireless communication devices and may use metadata of the transmission waveform that identifies the wireless communication device to associate the receiver neural network indicator from the wireless communication device with the transmission waveform. In some embodiments, in optional block 1014, the receiver, base station, decoder, and / or receiver neural network may select a receiver neural network for demodulating the transmission waveform.

[0140] In block 1016, the receiver, receiving wireless communication device, and / or decoder may use the wireless communication device's receiver neural network to reconstruct the time domain data signal from the received transmission waveform, as described in reference to Figure 13 1006) and a boosting neural network trained to reduce the PAPR in a signal for transmission by a transmitting wireless communication device. The receiver neural network in the receiver wireless communication device may be configured to demodulate the time domain transmission waveform to reconstruct the time domain data signal. By training based on output from a transmitter including the PRT neural network set and the boosting neural network, the receiver neural network is able to use information inherent in the PRT to more accurately recover the time domain data signal than is possible using conventional demodulation circuitry.

[0141] In block 1018 , a fast Fourier transform component (eg, 602 ) of the receiver wireless communication device may transform the reconstruction of the time-domain data signal to the frequency domain, thereby generating a reconstruction of the frequency-domain data tones that is output in block 1020 .

[0142] Figure 10B An exemplary method 1050 is shown that may be performed by a transmitter wireless communication device for encoding frequency domain data tones into a transmission waveform using machine learning based encoding for nonlinearity mitigation using PRT in accordance with various embodiments. Figure 1-10BThe method 1050 may be implemented in various components of a wireless communication device (e.g., base stations 110a-110d, 350, 402, wireless devices 120a-120e, 200, 320, 404, transmitter 700). To encompass alternative configurations implemented in various embodiments, the hardware implementing the method 1050 is generally referred to as a "transmitter wireless communication device."

[0143] In block 1002, the wireless communication device may receive frequency domain tones for transmission, as described for like numbered blocks of method 1000. In block 1004, the transmitter wireless communication device may transform the frequency domain data tones into time domain data signals, as described for like numbered blocks of method 1000.

[0144] In block 1052, the transmitter wireless communication device may generate a time-domain PRT using the time-domain data signal using a set of PRT neural networks (e.g., 502a, 502b, etc.) that have been trained in conjunction with a boosting neural network and a receiver neural network (e.g., 604) as described herein.

[0145] In block 1054, the transmitter wireless communication device may generate an output of the enhanced neural network based on an input of a final combined time domain signal, the final combined time domain signal comprising a time domain PRT combined with a previous combined time domain signal. Some embodiments may further include the transmitter wireless communication device generating a real feature map of the final combined time domain signal and a real feature map of a function of the final combined time domain signal, and in such embodiments, the transmitter wireless communication device may generate an output of the enhanced neural network based on the input of the real feature map of the final combined time domain signal and the real feature map of the function of the final combined time domain signal. Some embodiments may further include the transmitter wireless communication device generating a complex feature map of the final combined time domain signal and a complex feature map of the function of the final combined time domain signal, and in such embodiments, the transmitter wireless communication device may generate an output of the enhanced neural network based on the input of the complex feature map of the final combined time domain signal and the complex feature map of the function of the final combined time domain signal.

[0146] In block 1056, the transmitter wireless communication device may generate a time-domain wireless transmission waveform comprising the output of the enhanced neural network combined with the final combined time-domain signal.

[0147] Figure 11 An example of a method 1100 for training a machine learning neural network for generating a PRT based on input of data tones at the transmitter side and extracting the PRT and reconstructing the data tones at the receiver side is shown in accordance with various embodiments. Figure 1-11, method 1100 may be implemented in a wireless communication device (e.g., base stations 110a-110d, 350, 402, wireless devices 120a-120e, 200, 320, 404, transmitter 700, receiver 702) having a neural network (e.g., 502, 602) in or coupled to a processor (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432), or in a combination of a software-configured processor and dedicated hardware (such as a processor (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432, encoder 500, decoder 600) and various memory / cache controllers executing software within a peak reduction encoder and / or decoder system including other separate components). To encompass alternative configurations implemented in various embodiments, the hardware implementing method 1100 is referred to herein as a "training device."

[0148] In block 1102, a training device may receive frequency-domain data tones representing data signals that a transmitter may send to a receiver during wireless communication. In some embodiments, an encoder and / or a set of PRT neural networks (e.g., the set of PRR neural networks 502a, 502b) may receive the frequency-domain data tones in block 1102. In some embodiments, an encoder, an inverse fast Fourier transform component (e.g., the inverse fast Fourier transform component 506), and / or a processor may receive the frequency-domain data tones in block 1102.

[0149] In block 1104, the training device may tonally transform the frequency domain data into a time domain data signal for use in generating a transmission waveform for amplification and transmission. In some embodiments, in block 1104, an encoder, an inverse fast Fourier transform component, and / or a processor may tonally transform the frequency domain data into a time domain data signal.

[0150] In block 1106, the training device may generate a transmission waveform using the set of PRT neural networks (e.g., the set of PRR neural networks 502a, 502b) and the boosted neural network (e.g., the boosted neural network 508) of the transmitter wireless communication device, as described herein with reference to Figure 12As further described in method 1200 of

[0066] , the training device may process the transformed time-domain data signal in block 1104 via a set of PRT neural networks and a boosted neural network to generate a transmission waveform. As described herein, the set of PRT neural networks and the boosted neural network may be trained based on the accuracy of reconstructing the frequency-domain data tones by a trained receiver neural network (e.g., receiver neural network 604) and / or based on the deviation of acceptable transmissions from a distortion function applied to the transmission waveform. The set of PRT neural networks and the boosted neural network may be trained to reduce the PAPR in a signal for transmission by a transmitting wireless communication device. By generating a PRT for the frequency-domain data signal via the set of PRT neural networks and the boosted neural network and including the PRT in the transmission waveform, the transmission waveform may include information inherent in the PRT that may assist the receiver neural network in more accurately recovering the time-domain data signal than would be possible using conventional demodulation circuitry. In some embodiments, in block 1116, the encoder, the set of PRT neural networks, the boosted neural network, the signal summer (e.g., 504a, 504b, 520), and / or the processor may generate the transmission waveform.

[0151] In block 1108, the training device may distort the transmission waveform using a nonlinear function that models the power amplifier and propagation channel. The training device may distort the transmission waveform to simulate the conditions under which the transmission waveform is transmitted from the transmitter wireless communication device to the receiver wireless communication device. Including these distortions during training of the PRT neural network set, the boosted neural network, and the receiver neural network may train the neural network to account for the effects of the distortion on the transmission waveform and reconstruction of the frequency domain data tones. When applied to the transmitter wireless communication device and the receiver wireless communication device, training with the distortion of the transmission waveform may make the neural network more resistant to the effects of the distortion of the transmission waveform. In some embodiments, the processor of the training device may be configured to distort the transmission waveform in block 1108.

[0152] In block 1110, the training device may add noise to the transmission waveform. The training device may add noise to the transmission waveform to simulate interference during transmission of the transmission waveform from the transmitter wireless communication device to the receiver wireless communication device. Including the added noise during training of the PRT neural network set, the boosted neural network, and the receiver neural network may train the neural network to account for the effects of interference on the reconstruction of the transmission waveform and frequency-domain data tones. When applied to the transmitter wireless communication device and the receiver wireless communication device, training with noise added to the transmission waveform may make the neural network more resistant to the effects of interference during transmission of the waveform. In some embodiments, in block 1110, the processor of the training device may be configured to add noise to the transmission waveform.

[0153] In block 1112, the training device may use a receiver neural network to reconstruct a time domain data signal from the transmitted waveform, as described in reference Figure 13 1106, 1108, 1110 to generate a reconstruction of the time domain data signal. As described herein, the receiver neural network in the receiver can be trained based on outputs from a set of PRT neural networks and a boosting neural network (i.e., the set of PRT neural networks and the boosting neural network used at block 1106), which are trained to reduce the PAPR in the signal for transmission by the transmitting wireless communication device. The receiver neural network can be configured to demodulate the time domain transmission waveform to reconstruct the time domain data signal. By training based on outputs from the set of PRT neural networks and the boosting neural network, the receiver neural network can use information inherent in the PRT to more accurately recover the time domain data signal (compared to what is possible using conventional demodulation circuitry).

[0154] In block 1114, the training device may transform the reconstruction of the time domain data signal into the frequency domain, thereby generating a reconstruction of the frequency domain data tone. In some embodiments, in block 1114, the decoder, the fast Fourier transform component (e.g., 602), and / or the processor may transform the reconstruction of the time domain data signal into the frequency domain.

[0155] In block 1116, the training device may determine the error for the reconstruction of the frequency domain data tones and the transmission at the distortion function used to distort the transmission waveform in block 1108. The training device may be configured to compare the frequency domain data tones and the reconstruction of the frequency domain data tones by various known means to determine an error value (error1) for the reconstruction of the data tones. For example, the reconstruction of the frequency domain data tones and the quality and / or content of the frequency domain data tones may be compared. As a further example, the result of processing the reconstruction of the frequency domain data tones may indicate an error value for the reconstruction of the frequency domain data tones, and / or a comparison of the result with an expected result may indicate an error value for the reconstruction of the frequency domain data tones. In some embodiments, the training device may determine the error for the reconstruction of the frequency domain data tones by calculating the mean square error between the frequency domain data tones and the reconstruction of the frequency domain data tones. The training device may be configured to determine the error (error2) output by the distortion function based on the energy outside the allocated frequency band. In some embodiments, the total error can be obtained by a weighted average of the error (error1) from the comparison of the reconstruction of the frequency domain data tone with the frequency domain data tone and the error (error2) from the distortion function output. For example, for a certain "α" that can be found during hyperparameter tuning, the total error can be = error1 + α * error2. In some embodiments, in block 1116, the encoder, decoder, and / or processor can determine the error for the reconstructed data tone.

[0156] In block 1118, the training device may use the calculated errors to train the PRT neural network set, the boosting neural network, and the receiver neural network in conjunction with one another. The calculated errors may include any of the following: an error from a comparison of the reconstruction of the frequency-domain data tones with the frequency-domain data tones; an error from the output of a distortion function; or a weighted average of an error from a comparison of the reconstruction of the frequency-domain data tones with the frequency-domain data tones and an error from the output of a distortion function. The training device may be configured to update weight values ​​of the PRT neural network set and the boosting neural network used to generate the time-domain PRT and / or weight values ​​of the receiver neural network used to demodulate the transmission waveform to reconstruct the frequency-domain data tones (e.g., derived from a demodulated reconstruction of the time-domain data signal) to reduce the errors. The training device may use an algorithm configured to use the error values ​​and weights as inputs and output updated weights. In some embodiments, in block 1118, the encoder, decoder, and / or processor may use the errors for the reconstruction of the frequency-domain data tones to train the PRT neural network set, the boosting neural network, and the receiver neural network.

[0157] In some embodiments, method 1100 can be repeated to train the PRT neural network set, the boosting neural network, and the receiver neural network in combination until the difference between the frequency-domain data tones input to the transmitter using its PRT neural network and the frequency-domain reconstructed data tones output by the receiver neural network falls within an acceptable error threshold. In some embodiments, method 1100 can be repeated using a plurality of different frequency-domain data tones for multiple iterations of method 1100. Method 1100 can be repeated using a plurality of frequency-domain data tones until a number of the plurality of frequency-domain data tones (up to all of the plurality of frequency-domain data tones) result in an error value that does not exceed the error threshold. Successive iterations of method 1100 can use updated weights resulting from training the PRT neural network set, the boosting neural network, and / or the receiver neural network using the error values ​​in block 1118.

[0158] Various embodiments improve the operation of a wireless communication device transmitter and receiver compared to conventional transmitter / receiver pairs. By training a receiver neural network based on the output of a transmitter that uses a trained PRT neural network to generate a PRT, the trained receiver neural network is able to utilize information related to the PRT embedded in the transmission waveform output by the transmitter to more accurately (i.e., with less error) reconstruct the data tones (compared to what is possible using conventional demodulation circuitry that ignores the PRT). By training the PRT neural network and the receiver neural network in combination, a tight connection between the encoder and decoder can be coupled via the PRT inserted in the transmitted signal, thereby reducing the PAPR on the transmitter side to within acceptable levels, while also enabling better reconstruction of the data tones on the receiver side (compared to what is possible using conventional PRT generation circuitry and demodulation circuitry).

[0159] Figure 12 An example of a method for implementing a transmission waveform generation from data tones using a neural network according to various embodiments is shown. Figure 1-12, method 1200 can be implemented in a computing device (e.g., base stations 110a-110d, 350, 402, wireless devices 120a-120e, 200, 320, 404, transmitter 700, receiver 702), in general-purpose hardware, in dedicated hardware, in software executed in a processor (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432), or in a combination of a software-configured processor and dedicated hardware (such as a processor (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432, encoder 500, decoder 600) and various memory / cache controllers) executing software within a peak reduction encoder and / or decoder system that includes other separate components. To encompass alternative configurations implemented in various embodiments, the hardware implementing method 1200 is referred to herein as a "wireless communication device" and / or a "training device." In some embodiments, method 1200 may also describe block 1006 of method 1000 described herein with reference to FIG. 10 and block 1007 of method 1000 described herein with reference to FIG. Figure 11 Block 1106 of method 1100 is described.

[0160] In block 1202, the wireless communication device and / or the training device may receive a time domain data signal. The time domain data signal may be received in block 1004 of method 1000 as described with reference to FIG. 10 and / or as described with reference to FIG. Figure 11 The time domain data signal transformed in block 1104 of the described method 1100. In some embodiments, in block 1202, a processor, an encoder, and / or a PRT neural network (eg, the first PRT neural network 502a) may receive the time domain data signal.

[0161] In block 1204, the wireless communication device and / or training device may determine a time-domain PRT using the time-domain data signal. The time-domain data signal may be the time-domain data signal received in block 1202. In some embodiments, a PRT neural network may receive the time-domain data signal as input, process the time-domain data signal using weights of the PRT neural network, and generate a time-domain PRT as an inference. In some embodiments, the PRT neural network may process the time-domain data signal using a conventional PRT calculation algorithm, wherein the PRT neural network calculates one iteration of the conventional PRT calculation algorithm. In conventional PRT calculation algorithms, each iteration has the same weight values. However, the iteration of the conventional PRT calculation algorithm used in each PRT neural network may have different weight values. Training the PRT neural network may determine weight values ​​that may be different for each PRT neural network in generating the time-domain PRT, resulting in a more accurate reconstruction of the frequency-domain data tone (compared to when determined using a conventional PRT calculation algorithm (where the same weight values ​​are used for each iteration)). In some embodiments, in block 1204 , a processor, an encoder, and / or a PRT neural network (eg, the first PRT neural network 502 a ) may determine a time-domain PRT using the time-domain data signal.

[0162] In block 1206, the wireless communication device and / or the training device may combine the time-domain data signal and a time-domain PRT generated using the time-domain data signal to generate a time-domain combined signal. The first time-domain combined signal may be the result of combining the time-domain data signal and a time-domain PRT generated by a PRT neural network, such as a first PRT neural network, using the time-domain data signal as input. In some embodiments, the time-domain data signal and the time-domain PRT may be added together. In some embodiments, in block 1206, the processor, encoder, and / or signal summer (e.g., first signal summer 504a) may combine the time-domain data signal and the time-domain PRT to generate a time-domain combined signal.

[0163] In block 1208, the wireless communication device and / or the training device may determine a time-domain PRT using the time-domain combined signal. The time-domain combined signal may be the first time-domain combined signal generated in block 1206. In some embodiments, a PRT neural network may receive the time-domain combined signal as input, process the time-domain combined signal using weights of the PRT neural network, and generate a time-domain PRT as an inference. In some embodiments, the PRT neural network may process the time-domain combined signal using a conventional PRT calculation algorithm, wherein the PRT neural network calculates one iteration of the conventional PRT calculation algorithm. In conventional PRT calculation algorithms, each iteration has the same weight values. However, the iteration of the conventional PRT calculation algorithm used in each PRT neural network may have different weight values. Training the PRT neural network may determine weight values ​​that may be different for each PRT neural network generating the time-domain PRT, resulting in a more accurate reconstruction of the pitch of the frequency-domain data (compared to when determined by an unmodified version of the conventional PRT calculation algorithm (where the same weight values ​​are used for each iteration)). In some embodiments, in block 1208 , the processor, encoder, and / or PRT neural network (eg, second PRT neural network 502 b , third PRT neural network 502 b ) may determine a time-domain PRT using the time-domain combined signal.

[0164] In block 1210, the wireless communication device and / or the training device may combine the time-domain combined signal and the time-domain PRT generated using the time-domain combined signal to generate a successive time-domain combined signal. The successive time-domain combined signal may be the result of a combination of a previous time-domain combined signal (such as the first time-domain combined signal) and a time-domain PRT generated by a PRT neural network (such as the second PRT neural network) using the previous time-domain combined signal as input. In some embodiments, the previous time-domain combined signal and the time-domain PRT may be added together. In some embodiments, in block 1210, the processor, the encoder, and / or the signal summer (e.g., the second signal summer 504b) may combine the time-domain data signal and the time-domain PRT to generate a time-domain combined signal.

[0165] In some embodiments, blocks 1208 and 1210 may iteratively loop for any remaining PRT neural networks and signal summers. In some embodiments, method 1200 may not proceed to block 1212 until the final iteration of block 1210 has been performed. For example, the final iteration of block 1210 may be performed by a processor, an encoder, and / or a final signal summer.

[0166] In block 1212, the wireless communication device and / or the training device may generate an output of the enhanced neural network (e.g., enhanced neural network 508). The enhanced neural network may generate an output as described in reference Figure 14In some embodiments, at block 1212, the processor, encoder, and / or enhanced neural network may generate an output of the enhanced neural network.

[0167] In block 1214, the wireless communication device and / or the training device may output a transmission waveform. The transmission waveform may be a time-domain combined signal generated by the final iteration of combining the time-domain data signal and the time-domain PRT combined with the output of the enhanced neural network in block 1210. In some embodiments, in block 1214, the processor, encoder, and / or signal summer may output the transmission waveform.

[0168] Figure 13 An example of a method for reconstructing a time domain data signal from a transmission waveform using a neural network according to various embodiments is shown. Figure 1-13 , method 1300 can be implemented in a computing device (e.g., base stations 110a-110d, 350, 402, wireless devices 120a-120e, 200, 320, 404, transmitter 700, receiver 702), in general-purpose hardware, in dedicated hardware, in software executed in a processor (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432), or in a combination of a software-configured processor and dedicated hardware (such as a processor (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432, encoder 500, decoder 600) and various memory / cache controllers) executing software within a peak reduction encoder and / or decoder system that includes other separate components. To encompass alternative configurations implemented in various embodiments, the hardware implementing method 1300 is referred to herein as a "wireless communication device" and / or a "training device." In some embodiments, method 1300 may be included in block 1016 of method 1000 as described with reference to FIG. 10 and in block 1017 of method 1000 as described with reference to FIG. Figure 11 The operations performed in block 1112 of method 1100 are described.

[0169] In block 1302, the wireless communication device and / or training device may equalize the transmission waveform to generate an equalized time domain signal. Equalizing the transmission waveform may eliminate the effect of the propagation channel on the frequency response of the transmission waveform. Equalizing the transmission waveform may generate an equalized time domain signal (y n ) (e.g., equalized time-domain signals 624, 628). Equalization of the transmission waveform may be achieved using any known equalization process. In some embodiments, at block 1302, the processor, decoder, and / or channel equalizer component (e.g., channel equalizer component 610) may equalize the transmission waveform to generate an equalized time-domain signal.

[0170] In block 1304, the wireless communication device and / or training device may extract a time-domain PRT from the equalized time-domain signal. The time-domain PRT (pRT) may be extracted from the equalized time-domain signal (e.g., the equalized time-domain signal 624). n In some embodiments, in block 1304, the processor, decoder, and / or time-domain PRT extractor component (eg, time-domain PRT extractor component 612) may extract the time-domain PRT from the equalized time-domain signal.

[0171] In block 1306, the wireless communication device and / or the training device may generate a feature map based on the equalized time domain signal and the extracted time domain PRT. A feature map of the equalized time domain signal (e.g., a feature map of the equalized time domain signal 620) may be generated, such as two real feature maps or one complex feature map. A feature map of a function of the equalized time domain signal (e.g., a feature map of the equalized time domain signal 622) may be generated, such as a function y. n |y n | 2 A feature map of the time-domain PRT extracted from the equalized time-domain signal (e.g., a feature map of the time-domain PRT 626) may be generated, such as two real feature maps or one complex feature map. In some embodiments, in block 1306, the processor, decoder, channel equalizer component, and / or time-domain PRT extractor component may generate a feature map based on the equalized time-domain signal and the extracted time-domain PRT.

[0172] In block 1308, the wireless communication device and / or the training device may input the feature map into a receiver neural network (e.g., receiver neural network 604). The receiver neural network may receive as input the feature map of the equalized time domain signal, the feature map of the function of the equalized time domain signal, and the feature map of the time domain PRT. In some embodiments, the feature map may be input into a first layer of the receiver neural network (e.g., first neural network layer 614a). In some embodiments, the first layer of the receiver neural network may be a one-dimensional convolutional neural network layer with a 3-channel kernel. In some embodiments, in block 1308, the processor, the decoder, the channel equalizer component, the time domain PRT extractor component, the receiver neural network, and / or the first layer of the receiver neural network may input the feature map into the receiver neural network.

[0173] In block 1310, the wireless communication device and / or training device may use a receiver neural network to generate a demodulated transmission waveform. A first neural network layer may process the input feature map and output a first activation. The first activation output by the first neural network layer may be input to a second neural network layer (e.g., second neural network layer 616a) (e.g., a residual neural network (ResNet) block with a 3-channel kernel). The second neural network layer may process the input first activation and output a second activation. The second activation output by the second neural network layer may be input to a third neural network layer (e.g., third neural network layer 616b) (e.g., a ResNet block with a 3-channel kernel). The third neural network layer may process the input second activation and output a third activation. The third activation output by the third neural network layer may be input to a fourth neural network layer (e.g., fourth neural network layer 614b) (e.g., a 1-dimensional convolutional neural network layer with a 1-channel kernel). The fourth neural network layer may process the input third activation and output a demodulated transmission waveform in the time domain. In some embodiments, in block 1310, the processor, decoder, receiver neural network, and / or first, second, third, and / or fourth layers of the receiver neural network may use the receiver neural network to generate a demodulated transmit waveform.

[0174] In block 1312, the wireless communication device and / or the training device may combine the demodulated transmission waveform and the equalized time domain signal to generate a time domain reconstruction of the data signal. The demodulated transmission waveform and the equalized time domain signal (e.g., equalized time domain signal 628) output by the fourth neural network layer may be input to a signal adder (e.g., signal adder 618). In some embodiments, the demodulated transmission waveform and the equalized time domain signal may be added. The result of combining the demodulated transmission waveform and the equalized time domain signal may be a time domain reconstruction of the data signal. In some embodiments, in block 1312, the processor, the decoder, and / or the first, second, third, and / or signal adders may combine the demodulated transmission waveform and the equalized time domain signal to generate a time domain reconstruction of the data signal.

[0175] Figure 14 An example of a method for generating a transmission waveform from data tones using a neural network in accordance with some embodiments is shown. Figure 1-14, method 1400 can be implemented in a computing device (e.g., base stations 110a-110d, 350, 402, wireless devices 120a-120e, 200, 320, 404, transmitter 700, receiver 702), in general-purpose hardware, in dedicated hardware, in software executed in a processor (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432), or in a combination of software-configured processors and dedicated hardware (such as processors executing software within a peak reduction encoder and / or decoder system including other separate components (e.g., processors 210, 212, 214, 216, 218, 252, 256, 260, 428, 432, encoder 500, decoder 600) and various memory / cache controllers). To encompass alternative configurations implemented in various embodiments, the hardware implementing method 1400 is referred to herein as a "wireless communication device" and / or a "training device." In some embodiments, method 1400 may also be described in reference to Figure 12 The operations performed in block 1212 of method 1200 are described.

[0176] In block 1402, the wireless communication device and / or the training device may receive a final combined time domain signal. The final combined time domain signal may be obtained from a reference signal such as Figure 12 The final combination of the time-domain combined signal and the time-domain PRT in block 1210 of the described method 1200 produces a combined time-domain signal. In some embodiments, in block 1402, a processor, an encoder, and / or a boosted neural network (e.g., boosted neural network 508) may receive the final combined time-domain signal.

[0177] In block 1404, the wireless communication device and / or the training device may generate a feature map for the final combined time domain signal and a function of the final combined time domain signal. n ) can be used to generate a feature map (eg, feature map 516) of the final combined time domain signal, such as a real feature map and a complex feature map. The final combined time domain signal (x n ) can be used to generate a feature map (e.g., feature map 518) of a function of the final combined time-domain signal, such as the function x n |x n | 2 In some embodiments, in block 1404, the processor, the encoder, the enhanced neural network, and / or the feature map generator (e.g., the feature map generator 510) may generate a feature map for the final combined time-domain signal and a function of the final combined time-domain signal.

[0178] In block 1406, the wireless communication device and / or the training device may process the feature map in a first enhanced neural network layer (e.g., first neural network layer 512a) and generate an output of the first enhanced neural network layer. The feature map may be input to the first layer of the enhanced neural network, i.e., a one-dimensional convolutional neural network layer with a 3-channel kernel. The first layer of the enhanced neural network may process the feature map and output a first activation output resulting from the processing. In some embodiments, in block 1406, the processor, the encoder, the enhanced neural network, and / or the first layer of the enhanced neural network may process the feature map and generate an output.

[0179] In block 1408, the wireless communication device and / or training device may process the outputs of the enhanced neural network layers (e.g., the second neural network layer 514a, the third neural network layer 514b, the fourth neural network layer 514c, the fifth neural network layer 512b) and generate outputs of the enhanced neural network layers. The activation outputs of the previous layer of the enhanced neural network may be input to successive layers in the enhanced neural network. For example, the activation outputs of the first layer of the enhanced neural network may be input to the second layer of the enhanced neural network, such as a one-dimensional ResNet layer with a 3-channel kernel. Successive layers of the enhanced neural network may process the input activations and output successive activation outputs resulting from the processing. In some embodiments, in block 1408, the processor, the encoder, the enhanced neural network, and / or the second, third, fourth, and / or fifth layers of the enhanced neural network may process the outputs of the enhanced neural network layers and generate outputs.

[0180] In some embodiments, block 1408 may iteratively loop for any remaining enhanced neural network layers. In some embodiments, method 1400 may not proceed to block 1410 until the final iteration of block 1408 is performed. For example, the final iteration of block 1408 may be performed by a processor, an encoder, an enhanced neural network, and / or a fifth layer of the enhanced neural network.

[0181] In block 1410, the wireless communication device and / or training device may generate a time domain transmission waveform. A signal adder (e.g., signal adder 520) may combine the output of the final neural network layer with the final combined time domain signal to generate a time domain transmission waveform to be sent to the receiver. In some embodiments, the signal adder may add the output of the final neural network layer and the final combined time domain signal to generate the time domain transmission waveform. In some embodiments, in block 1410, the processor, encoder, enhanced neural network, and / or signal adder may generate the time domain transmission waveform.

[0182] Figure 15 An example of a transmission waveform 1500 with a time domain data signal and a time domain PRT on orthogonal subcarriers 1502, 1504 is shown in accordance with various embodiments. Figure 1-15 , an encoder (e.g., encoder 500) at a transmitter (e.g., transmitter 700) can generate a transmission waveform 1500 based on a time domain data signal and a time domain PRT generated from frequency domain data tones by a set of PRT neural networks (e.g., a set of PRT neural networks 502a, 502b) and a boosting neural network (e.g., boosting neural network 508). A signal adder (e.g., 520) of the transmitter (e.g., 700) can combine the time domain data signal on the data tone subcarrier 1502 and the time domain PRT on the orthogonal subcarrier 1504 through frequency division multiplexing. The time domain data signal and the time domain PRT can be sent by the transmitter on the orthogonal subcarriers 1502, 1504. The time domain PRT can be sent on a subcarrier 1504 that is orthogonal to the subcarrier 1502 reserved for the data tone.

[0183] Various embodiments (including methods 900, 1000, 1100, 1200, 1300, and 1400) may be performed in various network computing devices (e.g., in base stations 110a-110d, 350, 402). Figure 16 An example of a network computing device is shown in FIG. Figure 16 An example of a network computing device 1600 is shown. Figure 1-15 , the network computing device 1600 acts as a network element (such as a base station) of a communication network. The network computing device 1600 may include a processor 1601 coupled to a volatile memory 1602 and a large-capacity non-volatile memory (such as a disk drive 1603). The network computing device 1600 may also include a peripheral memory access device, such as a floppy disk drive, a compact disc (CD), or a digital video disc (DVD) drive 1606 coupled to the processor 1601. The network computing device 1600 may also include a network access port 1604 (or interface) coupled to the processor 1601 for establishing a data connection with a network (such as the Internet or a local area network coupled to other system computers and servers). The network computing device 1600 may include one or more antennas 1607 that can be connected to a wireless communication link for sending and receiving electromagnetic radiation. The network computing device 1600 may include additional access ports, such as USB, Firewire, Thunderbolt, etc., for coupling to peripheral devices, external memory, or other devices.

[0184] Various embodiments (including methods 900, 1000, 1100, 1200, 1300, and 1400) may be performed in various wireless devices (e.g., wireless devices 120a-120e, 200, 320, 404), Figure 17 An example of this is shown in Figure 17An example of a wireless device 1700 suitable for use with various embodiments is shown. Figure 1-15 , the wireless device 1700 may include a first SOC 202 (e.g., a SOC-CPU) coupled to a second SOC 204 (e.g., a 5G-capable SOC). The first SOC 202 and the second SOC 204 may be coupled to internal memory 426, 1716, a display 1712, and a speaker 1714. In addition, the wireless device 1700 may include an antenna 1704 for transmitting and receiving electromagnetic radiation, which may be connected to a wireless data link and / or cellular telephone transceiver 266, which is coupled to one or more processors in the first SOC 202 and / or the second SOC 204. The wireless device 1700 may also include a menu selection button or rocker switch 1720 for receiving user input.

[0185] The wireless device 1700 may also include a sound coding / decoding (CODEC) circuit 1710 that digitizes the sound received from the microphone into data packets suitable for wireless transmission and decodes the received sound packets to generate analog signals, which are provided to the speaker to generate sound. In addition, one or more of the processors in the first SOC 202 and the second SOC 204, the wireless transceiver 266, and the CODEC 1710 may include a digital signal processor (DSP) circuit (not shown separately).

[0186] The processors of the network computing device 1600 and the wireless device 1700 can be any programmable microprocessor, microcomputer, or one or more multi-processor chips that can be configured by software instructions (applications) to perform various functions, including the functions of the various embodiments described below. In some mobile devices, multiple processors may be provided, such as one processor within the SOC 204 dedicated to wireless communication functions and one processor within the SOC 202 dedicated to running other applications. Before accessing software applications and loading them into the processor, they can be stored in the memory 426, 1602, 1603, 1716. The processor may include internal memory sufficient to store the application software instructions.

[0187] Implementation examples are described in the following paragraphs. While some of the following implementation examples are described in terms of example methods, further example implementations may include: the example methods discussed in the following paragraphs implemented by a transmitter wireless communication device, the transmitter wireless communication device including a processing device configured with processor-executable instructions to perform the operations of the example methods; the example methods discussed in the following paragraphs implemented by a wireless computing device including means for performing the functions of the example methods; and the methods discussed in the following paragraphs implemented as a non-transitory processor-readable storage medium having processor-executable instructions stored thereon, the processor-executable instructions configured to cause a processor of the wireless computing device to perform the operations of the example methods.

[0188] Example 1. A method for reducing the peak-to-average power ratio of a wireless transmission waveform performed in a transmitter circuit of a wireless communication device, the method comprising: receiving a frequency domain data tone; transforming the frequency domain data tone into a time domain data signal; using a peak reduced tone (PRT) neural network set to generate a time domain PRT using the time domain data signal, wherein the PRT neural network set has been trained in conjunction with an enhancement neural network and a receiver neural network; generating an output of the enhancement neural network based on an input of a final combined time domain signal including the time domain PRT combined with a previous combined time domain signal; and generating a time domain wireless transmission waveform, the time domain wireless transmission waveform including the output of the enhancement neural network combined with the final combined time domain signal.

[0189] Example 2. The method according to Example 1 further includes: generating a real feature map of the final combined time domain signal and a real feature map of the function of the final combined time domain signal, wherein generating the output of the enhanced neural network based on the input of the final combined time domain signal includes: generating the output of the enhanced neural network based on the input of the real feature map of the final combined time domain signal and the real feature map of the function of the final combined time domain signal.

[0190] Example 3. The method according to Example 1 further includes: generating a complex feature map of the final combined time domain signal and a complex feature map of the function of the final combined time domain signal, wherein generating the output of the enhanced neural network based on the input of the final combined time domain signal includes: generating the output of the enhanced neural network based on the input of the complex feature map of the final combined time domain signal and the complex feature map of the function of the final combined time domain signal.

[0191] Example 4. The method of any one of Examples 1-3, further comprising: training the enhanced neural network using a weighted average of a first error and a second error, wherein the first error is a reconstruction from the frequency domain data tone and a comparison of the frequency domain data tone, and wherein the second error is a distortion function output based on energy outside the allocated frequency band.

[0192] Example 5. The method according to Example 4 further includes: using the weighted average of the first error and the second error to train the PRT neural network set.

[0193] Example 6. The method of any one of Examples 1-5, further comprising: determining whether the wireless communication device is configured to implement the enhanced neural network; and in response to determining that the wireless communication device is configured to implement the enhanced neural network, selecting a neural network implementation from a group of neural network implementations based on a modulation and coding scheme.

[0194] Example 7. The method according to any one of Examples 1-6 further includes: sending the time domain wireless transmission waveform to another wireless communication device having the receiver neural network.

[0195] Example 8. The method of any one of Examples 1-7, further comprising: receiving a neural network indicator from another wireless communication device having the receiver neural network, the neural network indicator being configured to indicate to the wireless communication device the enhanced neural network that has been trained together with the set of PRT neural networks and the receiver neural network; and selecting the enhanced neural network from a plurality of enhanced neural networks based on the neural network indicator.

[0196] Example 9. The method according to Example 8 further includes: receiving weights for the enhanced neural network from the other wireless communication device, wherein generating the time domain wireless transmission waveform including the output of the enhanced neural network combined with the final combined time domain signal includes: using the weights for the enhanced neural network to generate the time domain wireless transmission waveform.

[0197] As used in this application, the terms "component," "module," "system," and the like are intended to include computer-related entities such as, but not limited to, hardware, firmware, a combination of hardware and software, software, or software in execution, that are configured to perform a particular operation or function. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, a thread of execution, a program, and / or a computer. By way of illustration, both an application running on a wireless device and the wireless device can be referred to as a component. One or more components can be located within a process or thread of execution, and components can be localized on a processor or core or distributed across two or more processors or cores. In addition, these components can execute from various non-transitory computer-readable media having various instructions or data structures stored thereon. Components can communicate via local or remote processes, function or procedure calls, electronic signals, data packets, memory reads / writes, and other known communication methods associated with networks, computers, processors, or processes.

[0198] A variety of different cellular and mobile communication services and standards may be available or anticipated in the future, all of which can be implemented and benefit from various embodiments. Such services and standards include, for example, the Third Generation Partnership Project (3GPP), Long Term Evolution (LTE), third generation wireless mobile communication technology (3G), fourth generation wireless mobile communication technology (4G), fifth generation wireless mobile communication technology (5G), Global System for Mobile Communications (GSM), Universal Mobile Telecommunications System (UMTS), 3GSM, General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA) systems (such as cdmaOne, CDMA1020™), Enhanced Data Rates for GSM Evolution (EDGE), Advanced Mobile Phone System (AMPS), Digital AMPS (IS-136 / TDMA), Evolution Data Optimized (EV-DO), Digital Enhanced Cordless Telecommunications (DECT), Worldwide Interoperability for Microwave Access (WiMAX), Wireless Local Area Networks (WLAN), Wi-Fi Protected Access I and II (WPA, WPA2), and Integrated Digital Enhanced Network (iDEN). Each of these technologies involves, for example, the transmission and reception of voice, data, signaling, or content messages. It should be understood that, unless specifically recited in the language of the claims, any reference to terms or technical details related to individual telecommunication standards or technologies is for illustrative purposes only and is not intended to limit the scope of the claims to a particular communication system or technology.

[0199] The various embodiments shown and described are provided merely as examples to illustrate various features of the claims. However, the features shown and described with respect to any given implementation are not necessarily limited to the associated implementation and may be used or combined with other embodiments shown and described. Furthermore, the claims are not intended to be limited by any one example implementation. For example, one or more of the operations of the methods disclosed herein may be replaced by or combined with one or more operations of the methods disclosed herein.

[0200] As used herein, a phrase referring to "at least one of" a list of items refers to any combination of those items, including single members. As an example, "at least one of a, b, or c" is intended to encompass: a, b, c, ab, ac, bc, and abc.

[0201] The various illustrative logics, logical blocks, modules, components, circuits, and algorithmic operations described in conjunction with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or a combination of both. The interchangeability of hardware and software has been generally described and illustrated in the various illustrative components, blocks, modules, circuits, and processes described above in terms of functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system.

[0202] The hardware and data processing apparatus for implementing the various illustrative logic units, logic blocks, modules, and circuits described in conjunction with the various aspects disclosed herein may be implemented or performed using a general-purpose single-chip or multi-chip processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof, designed to perform the functions described herein. A general-purpose processor may be a microprocessor or any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. In some embodiments, specific processes and methods may be performed by circuits specific to a given function.

[0203] In one or more aspects, the functions described may be implemented in hardware, digital electronic circuitry, computer software, firmware (including the structures disclosed in this specification and their structural equivalents), or any combination thereof. Embodiments of the subject matter described in this specification may also be implemented as one or more computer programs, i.e., one or more modules of computer program instructions, encoded on computer storage media for execution by, or to control the operation of, data processing apparatus.

[0204] If implemented in software, the function can be stored as one or more instructions or codes on a computer-readable medium or transmitted through it. The process of the method or algorithm disclosed herein can be implemented in a processor-executable software module that can be located on a computer-readable medium. Computer-readable media include both computer storage media and communication media, and the communication media include any media that can be implemented to transfer a computer program from one place to another. The storage medium can be any available medium that can be accessed by a computer. By way of example and not limitation, such computer-readable media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage device, or any other medium that can be used to store the desired program code in the form of an instruction or data structure and can be accessed by a computer. In addition, any connection can be appropriately referred to as a computer-readable medium. As used herein, disks and optical disks include compact discs (CDs), laser discs, optical discs, digital versatile discs (DVDs), floppy disks and Blu-ray discs, wherein disks usually copy data magnetically, while optical discs use lasers to copy data optically. The combination of the above should also be included in the scope of computer-readable media. Additionally, the operations of a method or algorithm may reside as one or any combination or set of codes and instructions on a machine-readable medium or computer-readable medium, which may be incorporated into a computer program product.

[0205] Various modifications to the embodiments described in this disclosure may be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of this disclosure. Therefore, the claims are not intended to be limited to the embodiments shown herein, but are to be accorded the widest scope consistent with this disclosure, the principles and novel features disclosed herein.

[0206] In addition, those skilled in the art will readily appreciate that the terms "upper" and "lower" are sometimes used for convenience in describing the figures and indicate relative positions corresponding to the orientation of the figures on appropriately oriented pages, and may not reflect the correct orientation of the device being implemented.

[0207] In addition, certain features described in this specification in the context of separate embodiments may also be implemented in combination in a single implementation. Conversely, individual features described in the context of a single implementation may also be implemented in multiple embodiments individually or in any suitable subcombination. Furthermore, while features may be described above as acting in certain combinations, and even initially claimed as such, in some cases one or more features from a claimed combination may be removed from that combination, and a claimed combination may involve subcombinations or variations of subcombinations.

[0208] Similarly, although operations are depicted in a particular order in the figures, this should not be understood as requiring that such operations be performed or executed in the particular order shown or in the sequential order, or that all of the illustrated operations be performed, in order to achieve the desired result. In addition, the figures may schematically depict one or more example processes in the form of flow charts. However, other operations not depicted may be incorporated into the schematically illustrated example processes. For example, one or more additional operations may be performed before, after, simultaneously with, or between any of the illustrated operations. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of the various system components in the above-described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products. In addition, other embodiments are within the scope of the following claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve the desired result.

Claims

1. A method for reducing a peak-to-average power ratio of a wireless transmission waveform, performed in a transmitter circuit of a wireless communication device, the method comprising: receiving frequency domain data tones; Tone-transforming the frequency-domain data into a time-domain data signal; generating a time-domain PRT using the time-domain data signal using a set of peak-reduced pitch PRT neural networks, wherein the set of PRT neural networks has been trained in conjunction with a boosting neural network and a receiver neural network; generating an output of the enhanced neural network based on an input comprising a final combined time-domain signal of the time-domain PRT combined with a previous combined time-domain signal; as well as A time-domain wireless transmission waveform is generated, the time-domain wireless transmission waveform comprising the output of the enhanced neural network combined with the final combined time-domain signal.

2. The method according to claim 1, further comprising: generating a real number feature map of the final combined time domain signal and a real number feature map of a function of the final combined time domain signal, Wherein, generating the output of the enhanced neural network based on the input of the final combined time domain signal includes: generating the output of the enhanced neural network based on the input of the real feature map of the final combined time domain signal and the real feature map of the function of the final combined time domain signal.

3. The method according to claim 1, further comprising: generating a complex feature map of the final combined time domain signal and a complex feature map of a function of the final combined time domain signal, Wherein, generating the output of the enhanced neural network based on the input of the final combined time domain signal includes: generating the output of the enhanced neural network based on the input of the complex feature map of the final combined time domain signal and the complex feature map of the function of the final combined time domain signal.

4. The method according to claim 1, further comprising: The enhanced neural network is trained using a weighted average of a first error and a second error, wherein the first error is from reconstructing and comparing the frequency domain data tones, and wherein the second error is a distortion function output based on energy outside of an allocated frequency band.

5. The method according to claim 4, further comprising: The PRT neural network ensemble is trained using the weighted average of the first error and the second error.

6. The method according to claim 1, further comprising: determining whether the wireless communication device is configured to implement the enhanced neural network; as well as In response to determining that the wireless communication device is configured to implement the enhanced neural network, a neural network implementation is selected from a group of neural network implementations based on a modulation and coding scheme.

7. The method according to claim 1, further comprising: The time-domain wireless transmission waveform is transmitted to another wireless communication device having the receiver neural network.

8. The method according to claim 1, further comprising: receiving a neural network indicator from another wireless communication device having the receiver neural network, the neural network indicator configured to indicate to the wireless communication device the enhanced neural network that has been trained with the set of PRT neural networks and the receiver neural network; as well as The enhanced neural network is selected from a plurality of enhanced neural networks based on the neural network indicator.

9. The method according to claim 8, further comprising: weights for the enhanced neural network are received from the other wireless communication device, wherein generating the time-domain wireless transmission waveform comprising the output of the enhanced neural network combined with the final combined time-domain signal comprises generating the time-domain wireless transmission waveform using the weights for the enhanced neural network.

10. A transmitter wireless communication device comprising: A processing device configured with processor-executable instructions to perform operations comprising: receiving frequency domain data tones; Tone-transforming the frequency-domain data into a time-domain data signal; generating a time-domain PRT using the time-domain data signal using a set of peak-reduced pitch PRT neural networks, wherein the set of PRT neural networks has been trained in conjunction with a boosting neural network and a receiver neural network; generating an output of the enhanced neural network based on an input comprising a final combined time-domain signal of the time-domain PRT combined with a previous combined time-domain signal; and A time-domain wireless transmission waveform is generated, the time-domain wireless transmission waveform comprising the output of the enhanced neural network combined with the final combined time-domain signal.

11. The transmitter wireless communication device of claim 10, wherein: The processing device is configured with processor-executable instructions to perform operations further comprising: generating a real characteristic map of the final combined time-domain signal and a real characteristic map of a function of the final combined time-domain signal, Wherein, the processing device is configured with processor-executable instructions to perform operations such that generating the output of the enhanced neural network based on the input of the final combined time domain signal includes: generating the output of the enhanced neural network based on the input of the real feature map of the final combined time domain signal and the real feature map of the function of the final combined time domain signal.

12. The transmitter wireless communication device of claim 10, wherein: The processing device is configured with processor-executable instructions to perform operations further comprising: generating a complex characteristic map of the final combined time-domain signal and a complex characteristic map of a function of the final combined time-domain signal, Wherein, the processing device is configured with processor-executable instructions to perform operations such that generating the output of the enhanced neural network based on the input of the final combined time domain signal includes: generating the output of the enhanced neural network based on the input of the complex feature map of the final combined time domain signal and the complex feature map of the function of the final combined time domain signal.

13. The transmitter wireless communication device of claim 10, wherein: The processing device is configured with processor-executable instructions to perform operations further comprising: training the enhanced neural network using a weighted average of a first error and a second error, wherein the first error is from a reconstruction of the frequency-domain data tones and a comparison of the frequency-domain data tones, and wherein the second error is output by a distortion function based on energy outside of the allocated frequency band.

14. The transmitter wireless communication device of claim 13, wherein: The processing device is configured with processor-executable instructions to perform operations further comprising training the set of PRT neural networks using the weighted average of the first error and the second error.

15. The transmitter wireless communication device of claim 10, wherein: The processing device is configured with processor-executable instructions to perform operations further comprising: determining whether the transmitter wireless communication device is configured to implement the enhanced neural network; as well as In response to determining that the transmitter wireless communication device is configured to implement the enhanced neural network, a neural network implementation is selected from a group of neural network implementations based on a modulation and coding scheme.

16. The transmitter wireless communication device of claim 10, wherein: The processing device is configured with processor-executable instructions to perform operations further comprising: transmitting the time-domain wireless transmission waveform to another wireless communication device having the receiver neural network.

17. The transmitter wireless communication device of claim 10, wherein: The processing device is configured with processor-executable instructions to perform operations further comprising: receiving a neural network indicator from another wireless communication device having the receiver neural network, the neural network indicator configured to indicate to the transmitter wireless communication device the enhanced neural network that has been trained with the set of PRT neural networks and the receiver neural network; as well as The enhanced neural network is selected from a plurality of enhanced neural networks based on the neural network indicator.

18. The transmitter wireless communication device of claim 17, wherein: The processing device is configured with processor-executable instructions to perform operations further comprising: receiving weights for the enhanced neural network from the other wireless communication device, wherein generating the time-domain wireless transmission waveform comprising the output of the enhanced neural network combined with the final combined time-domain signal comprises: generating the time-domain wireless transmission waveform using the weights for the enhanced neural network.

19. A wireless computing device comprising: a unit for receiving frequency domain data tones; a unit for tone transforming the frequency domain data into a time domain data signal; means for generating a time-domain PRT using the time-domain data signal using a set of peak-reduced pitch PRT neural networks, wherein the set of PRT neural networks has been trained in conjunction with a boosting neural network and a receiver neural network; means for generating an output of the enhanced neural network based on an input comprising a final combined time-domain signal of the time-domain PRT combined with a previous combined time-domain signal; as well as Means for generating a time-domain wireless transmission waveform comprising the output of the enhanced neural network combined with the final combined time-domain signal.

20. The wireless computing device of claim 19, further comprising: a unit for generating a real-number characteristic map of the final combined time-domain signal and a real-number characteristic map of a function of the final combined time-domain signal, Among them, the unit for generating the output of the enhanced neural network based on the input of the final combined time domain signal includes: a unit for generating the output of the enhanced neural network based on the input of the real feature map of the final combined time domain signal and the real feature map of the function of the final combined time domain signal.

21. The wireless computing device of claim 19, further comprising: a unit for generating a complex feature map of the final combined time domain signal and a complex feature map of a function of the final combined time domain signal, Among them, the unit for generating the output of the enhanced neural network based on the input of the final combined time domain signal includes: a unit for generating the output of the enhanced neural network based on the input of the complex feature map of the final combined time domain signal and the complex feature map of the function of the final combined time domain signal.

22. The wireless computing device of claim 19, further comprising: Means for training the enhanced neural network using a weighted average of a first error and a second error, wherein the first error is from reconstructing and comparing the frequency domain data tones, and wherein the second error is a distortion function output based on energy outside of an allocated frequency band.

23. The wireless computing device of claim 22, further comprising: means for training the ensemble of PRT neural networks using the weighted average of the first error and the second error.

24. The wireless computing device of claim 19, further comprising: means for determining whether the wireless computing device is configured to implement the enhanced neural network; as well as Means for selecting a neural network implementation from a group of neural network implementations based on a modulation and coding scheme in response to determining that the wireless computing device is configured to implement the enhanced neural network.

25. The wireless computing device of claim 19, further comprising: Means for transmitting the time-domain wireless transmission waveform to another wireless computing device having the receiver neural network.

26. The wireless computing device of claim 19, further comprising: means for receiving a neural network indicator from another wireless computing device having the receiver neural network, the neural network indicator configured to indicate to the wireless computing device the enhanced neural network that has been trained with the set of PRT neural networks and the receiver neural network; as well as Means for selecting the enhanced neural network from a plurality of enhanced neural networks based on the neural network indicator.

27. The wireless computing device of claim 26, further comprising: means for receiving weights for the augmented neural network from the other wireless computing device, wherein means for generating the time-domain wireless transmission waveform comprising the output of the augmented neural network combined with the final combined time-domain signal comprises means for generating the time-domain wireless transmission waveform using the weights for the augmented neural network.

28. A non-transitory processor-readable medium having stored thereon processor-executable instructions configured to cause a processor of a wireless computing device to perform operations comprising: receiving frequency domain data tones; Tone-transforming the frequency-domain data into a time-domain data signal; generating a time-domain PRT using the time-domain data signal using a set of peak-reduced pitch PRT neural networks, wherein the set of PRT neural networks has been trained in conjunction with a boosting neural network and a receiver neural network; generating an output of the enhanced neural network based on an input comprising a final combined time-domain signal of the time-domain PRT combined with a previous combined time-domain signal; as well as A time-domain wireless transmission waveform is generated, the time-domain wireless transmission waveform comprising the output of the enhanced neural network combined with the final combined time-domain signal.

29. The non-transitory processor-readable medium of claim 28, wherein: The processor-executable instructions are configured to cause the processor of the wireless computing device to perform operations further comprising: generating a real characteristic map of the final combined time-domain signal and a real characteristic map of a function of the final combined time-domain signal, The processor-executable instructions are configured to cause the processor of the wireless computing device to perform operations such that generating the output of the enhanced neural network based on the input of the final combined time domain signal includes: generating the output of the enhanced neural network based on the input of the real feature map of the final combined time domain signal and the real feature map of the function of the final combined time domain signal.

30. The non-transitory processor-readable medium of claim 28, wherein: The processor-executable instructions are configured to cause the processor of the wireless computing device to perform operations further comprising: generating a complex characteristic map of the final combined time-domain signal and a complex characteristic map of a function of the final combined time-domain signal, The processor-executable instructions are configured to cause the processor of the wireless computing device to perform operations such that generating the output of the enhanced neural network based on the input of the final combined time domain signal includes: generating the output of the enhanced neural network based on the input of the complex feature map of the final combined time domain signal and the complex feature map of the function of the final combined time domain signal.

Images