System and method for mixing input data with coefficient data
The system addresses the inflexibility of existing wireless communication hardware by mixing input data with coefficient data through MAC and MLU layers, enabling efficient and adaptable processing across multiple wireless protocols, including 5G, with reduced hardware requirements and power consumption.
Patent Information
- Application Number
- CN202411127192.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2019-11-13
- Filing Date
- 2020-10-16
- Publication Date
- 2024-12-06
AI Technical Summary
Existing hardware platforms for wireless communications, particularly in 5G systems, require custom signal processing methods and lack flexibility in handling various wireless protocols, leading to long design times and inefficiencies.
A system and method for mixing input data with coefficient data using multiple layers of multiplication/accumulation units (MAC units) and memory lookup units (MLUs), allowing for configurable algorithms that adapt to different wireless protocols, such as 5G, and reducing the number of MAC units and MLUs while maintaining processing power.
Enables flexible and efficient processing of wireless communications across multiple protocols, including 5G, by using a power-efficient framework that operates with minimal power consumption and integrates seamlessly with various standards, reducing design time and optimizing hardware usage.
Smart Images

Figure CN119093961A_ABST
Abstract
Description
[0001] Information about divisional applications
[0002] This case is a divisional application. The parent case of the divisional application is an invention patent application with an application date of October 16, 2020, application number 202080079110.5, and invention name "System and method for mixing input data with coefficient data". Technical Field
[0003] The present description relates generally to wireless communications, for example, systems and methods for mixing input data with coefficient data. Background Art
[0004] Some hardware (e.g., silicon) computing platforms can be used to implement digital signal processing for wireless communications, such as digital baseband processing or digital front-end implementations. For example, multimedia processing and digital radio frequency (RF) processing can be implemented in a digital front-end implementation of a wireless transceiver such as implemented by an application-specific integrated circuit (ASIC). A variety of hardware platforms can implement such digital signal processing, such as an ASIC, a digital signal processor (DSP) implemented as part of a field programmable gate array (FPGA), or a system on a chip (SoC). However, each of these solutions often requires the implementation of a custom signal processing method specific to the hardware implementation. For example, a digital signal processor can implement a turbo decoding application for data in a custom design of an FPGA.
[0005] Additionally, there is a focus on evolving wireless communications to “fifth generation” (5G) systems. 5G promises increased speed and ubiquity, but the methods used to handle 5G wireless communications have not yet been fully determined. Summary of the invention
[0006] An example system is disclosed herein. In an embodiment of the present disclosure, a system includes at least one processing unit and a non-transitory computer-readable medium encoded with executable instructions, the executable instructions being configured to cause the system to perform operations including: obtaining a processing mode selection corresponding to a processing mode for the at least one processing unit; obtaining input data for a transmission in a radio frequency (RF) domain at a plurality of antennas based at least in part on the processing mode selection, each of the plurality of antennas receiving a portion of the transmission in the RF domain through a corresponding wireless channel in a plurality of wireless channels; mixing the input data and a delayed version of a corresponding output of the first MAC unit layer using a plurality of coefficients in a first multiplication / accumulation unit (MAC unit) layer of a plurality of MAC unit layers, the plurality of coefficients being specific to the processing mode selection; and providing output data based on mixing the input data using the plurality of coefficients, the output data representing a portion of the transmission processed according to the processing mode selection, wherein providing the output data includes providing the output data so that a wireless transceiver processing unit receives the output data for further processing of the transmission. Mixing the input data with the plurality of coefficients includes, for additional ones of the plurality of MAC unit layers, mixing respective processing results and delayed versions of at least a portion of the respective processing results using additional ones of the plurality of coefficients.
[0007] Additionally or alternatively, the number of the plurality of MAC unit layers corresponds to the number of wireless channels in the plurality of wireless channels.
[0008] Additionally or alternatively, after the first MAC unit layer, each subsequent layer of additional layers of the plurality of MAC units includes a smaller number of MAC units than a previous layer of additional layers of MAC units.
[0009] Additionally or alternatively, the second layer of the plurality of MAC units includes one less MAC unit than the number of MAC units in the first layer of MAC units, wherein the additional layer of MAC units includes the second layer of the plurality of MAC units.
[0010] Additionally or alternatively, each MAC unit in the first layer of MAC units receives input data and a respective delayed version of a respective output of the first layer of MAC units.
[0011] Additionally or alternatively, the operations include retrieving the plurality of coefficients from a memory database.
[0012] Additionally or alternatively, operations include configuring at least one processing unit to select a plurality of coefficients based on a processing mode selection.
[0013] Additionally or alternatively, operations include selecting a processing mode from a plurality of processing modes, each processing mode of the plurality of processing modes being associated with a respective plurality of coefficients.
[0014] Additionally or alternatively, each respective plurality of coefficients is stored in a cloud computing database configured to communicate with a non-transitory computer-readable medium.
[0015] Additionally or alternatively, the processing mode corresponds to at least one of a single processing mode, a multi-processing mode, or a full processing mode.
[0016] Additionally or alternatively, the single processing mode includes a fast Fourier transform (FFT) processing mode, an inverse fast Fourier transform (IFFT) processing mode, a decoding mode, a turbo decoding processing mode, a decoding processing mode, a Reed-Solomon processing mode, an interleaver processing mode, a deinterleaving processing mode, a modulation mapping processing mode, a demodulation mapping processing mode, a scrambling processing mode, a descrambling processing mode, or a channel estimation processing mode.
[0017] Additionally or alternatively, the modulation mapping processing mode is associated with at least one of GFDM, FBMC, UFMC, DFDM, SCMA, NOMA, MUSA, or FTN.
[0018] Additionally or alternatively, the multiple processing modes include at least two of a fast Fourier transform (FFT) processing mode, an inverse fast Fourier transform (IFFT) processing mode, a decoding mode, a turbo decoding processing mode, a decoding processing mode, a Reed-Solomon processing mode, an interleaver processing mode, a deinterleaving processing mode, a modulation mapping processing mode, a demodulation mapping mode, a scrambling mode, a descrambling mode, or a channel estimation mode.
[0019] Additionally or alternatively, the full processing mode includes a wireless transmitter processing mode or a wireless receiver processing mode.
[0020] Additionally or alternatively, the plurality of coefficients corresponds to a non-linear mapping of input data to output data, the non-linear mapping representing a portion of the transmission that is processed in accordance with the processing mode selection.
[0021] Additionally or alternatively, the nonlinear mapping includes a Gaussian function, a piecewise linear function, a sigmoid function, a thin plate spline function, a polyquadratic function, a cubic approximation, or an inverse polyquadratic function.
[0022] Additionally or alternatively, mixing the input data with the plurality of coefficients includes multiplying a portion of the input data with at least one of the plurality of coefficients and multiplying a delayed version of a corresponding output of the first MAC unit layer with at least another of the plurality of coefficients to produce a coefficient multiplication result; and accumulating the coefficient multiplication result to be further multiplied and accumulated with other portions of the input data and additional coefficients of the plurality of coefficients.
[0023] Additionally or alternatively, mixing the input data with the plurality of coefficients is included at a respective delay unit associated with the first MAC unit layer, delaying a respective output of the first MAC unit layer to generate a delayed version of the respective output of the first MAC unit layer.
[0024] Additionally or alternatively, mixing the corresponding processing result and the delayed version of at least a portion of the corresponding processing result using additional coefficients from the plurality of coefficients includes multiplying the corresponding processing result with at least one of the additional coefficients from the plurality of additional coefficients, and multiplying the delayed version of at least a portion of the corresponding processing result with at least another of the additional coefficients from the plurality of coefficients.
[0025] Additionally or alternatively, at least one processing unit includes a plurality of memory lookup units (MLUs) configured to store and provide respective coefficient multiplication results, wherein portions of the plurality of memory lookup units are configured to provide output data based on input data mixed using the plurality of coefficients.
[0026] Additionally or alternatively, the at least one processing unit further includes a plurality of delay units configured to provide delayed versions of respective outputs of the first MAC unit layer based on portions of respective coefficient multiplication results provided by respective MLUs of the plurality of MLUs.
[0027] Additionally or alternatively, at least one processing unit further includes an additional plurality of delay units configured to provide delayed versions of at least a portion of corresponding processing results, wherein at least a portion of the corresponding processing results corresponds to a portion of corresponding coefficient multiplication results of a plurality of MLUs associated with additional MAC unit layers in the plurality of MAC unit layers.
[0028] In another aspect of the present disclosure, the method includes: at a first multiplication / accumulation processing unit (MAC unit) layer of multiple MAC units, calculating input data and a delayed version of a corresponding output of the first MAC unit layer by multiple coefficients to generate a first processing result; at an additional MAC unit layer of the multiple MAC units, calculating the first processing result and a delayed version of at least a portion of the first processing result by an additional plurality of coefficients to generate a second processing result; providing output data based in part on the second processing result, the output data representing the input data processed according to aspects of a wireless protocol; and transmitting the output data via an RF antenna.
[0029] Additionally or alternatively, further comprising obtaining respective portions of the input data at a plurality of antennas, wherein each of the plurality of antennas receives the respective portion of the input data in an RF wireless domain via a respective wireless channel of a plurality of wireless channels.
[0030] Additionally or alternatively, the number of additional MAC unit layers is associated with the number of radio channels in the plurality of radio channels.
[0031] In another aspect of the present disclosure, an apparatus includes: a first stage of a combiner configured to receive wireless input data and evaluate at least one nonlinear function using a combination of the wireless input data and a delayed version of the combination of the wireless input data to provide intermediate data; and at least one second stage of a combiner configured to receive the intermediate data and combine the intermediate data using a set of predetermined weights to provide output data of the set of predetermined weights based at least in part on a wireless protocol associated with the wireless input data.
[0032] Additionally or alternatively, the set of predetermined weights is based on training of a recurrent neural network on delayed versions of known input data and combinations of known input data.
[0033] Additionally or alternatively, further included is a mode configuration control configured to select a set of weights for the recurrent neural network based on a wireless protocol associated with the wireless input data. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1A is a schematic diagram of a computing system arranged according to examples described herein.
[0035] Figure 1B is a schematic diagram of a neural network arranged according to the examples described herein.
[0036] Figure 1C is a schematic diagram of a recursive neural network arranged according to the examples described herein.
[0037] Figures 2A to 2C is a schematic diagram of processing units arranged in a system according to examples described herein.
[0038] Figure 3 is a schematic diagram of a wireless transmitter.
[0039] Figure 4 is a schematic diagram of a wireless receiver.
[0040] Figure 5 is a schematic diagram of a wireless transmitter according to examples described herein.
[0041] Figure 6 is a schematic diagram of a wireless transmitter according to examples described herein.
[0042] Figure 7 is a schematic diagram of a wireless transmitter according to examples described herein.
[0043] Figure 8is a flow chart of a method arranged according to examples described herein.
[0044] Fig. 9 is a flow chart of a method arranged according to examples described herein.
[0045] Fig.10 is a schematic diagram of a wireless communication system arranged in accordance with aspects of the present disclosure.
[0046] Fig.11 is a schematic diagram of another wireless communication system arranged according to aspects of the present disclosure. DETAILED DESCRIPTION
[0047] Currently there is a focus on evolving wireless communications to "fifth generation" (5G) systems. 5G is expected to increase speed and ubiquity, but the methods used to process 5G wireless communications have not yet been determined. The lead time for designing and processing hardware platforms for wireless communications can be long. Therefore, in some instances, it may be advantageous to design and / or process a hardware platform for 5G wireless communications that can process wireless communications using configurable algorithms - in this way, the algorithms used by the hardware platform may not need to be decided until after the platform is designed and / or manufactured.
[0048] Examples described herein include systems and methods, including wireless devices and systems having examples of mixing input data with coefficient data. The input data may be any data input for digital signal processing. The coefficient data may be any data specific to a wireless protocol. Examples of wireless protocols include, but are not limited to, 5G wireless systems utilizing wireless protocols, such as filter bank multi-carrier (FBMC), generalized frequency division multiplexing (GFDM), universal filtered multi-carrier (UFMC) transmission, dual orthogonal frequency division multiplexing (BFDM), sparse code multiple access (SCMA), non-orthogonal multiple access (NOMA), multi-user shared access (MUSA), and super-Nyquist (FTN) signaling with time-frequency packaging. In general, any wireless protocol including any 5G wireless protocol may be represented by the coefficient data disclosed herein. The input data may be mixed with the coefficient data to generate output data. For example, a computing system having a processing unit may mix input data for transmission in a radio frequency (RF) wireless domain with coefficient data to generate output data representing a transmission processed according to a wireless protocol in the RF wireless domain. In some examples, the computing system generates an approximation of the output data. For example, the output data may be an approximation of the output data generated when the input data is processed in hardware (e.g., an FPGA) that is specifically designed to implement the wireless protocol to which the coefficients correspond.
[0049] The examples described herein further include systems and methods, including wireless devices and systems having instances of mixing input data with such coefficient data in multiple layers of multiplication / accumulation units (MAC units) and corresponding memory lookup units (MLUs). For example, the number of MAC unit layers may correspond to the number of wireless channels, such as the number of channels received at corresponding antennas in multiple antennas. In addition, the number of MAC units and MLUs utilized is associated with the number of channels. For example, the second layer of MAC units and MLUs may include m-1 MAC units and MLUs, where m represents the number of antennas, each antenna receiving a portion of the input data. Advantageously, when utilizing such a hardware framework, the processing power of the generated output data can be maintained while reducing the number of MAC units and MLUs used for such processing in the electronic device. However, in some instances where board space may not be limited, a hardware framework including m MAC units and m MLUs in each layer may be utilized, where m represents the number of antennas.
[0050] Although the above examples of mixing input data with coefficient data have been described with respect to the RF wireless domain, it should be understood that wireless communication data may be processed from the perspective of different domains, such as the time domain (e.g., time division multiple access (TDMA)), the frequency domain (e.g., orthogonal frequency division multiple access (OFDMA), and / or the code domain (e.g., code division multiple access (CDMA)).
[0051] Advantageously, in some instances, the systems and methods described herein may operate in accordance with multiple standards and / or with multiple applications, including changes or upgrades to each of them; in contrast to the inflexible framework of ASIC-based solutions. In some instances, as discussed herein with respect to processing units implementing multiplication, addition, or accumulation functions, instances of the systems and methods described herein may operate on a power-efficient framework that consumes minimal power; in contrast to the extremely power-hungry framework of FPGA / DSP-based solutions. In some instances, the systems and methods described herein may operate using a substantially integrated framework from a unified programming language perspective; in contrast to the various programming languages required to integrate SoC solutions, which may present programming challenges when implementing heterogeneous interfaces of control units, computing units, data units, and accelerator units.
[0052] Examples described herein include systems and methods including wireless transmitters and receivers having examples of mixing input data with coefficient data. For example, digital signal processing aspects of a wireless transmitter may be implemented by mixing input data with coefficient data, as described herein. In this manner, a computing system may output data representing the operation of an RF front end to modulate input data for RF wireless transmission. In some examples, coefficient data may be mixed with input data to represent certain operations, such as: block coding the input data; interleaving block coded input data; mapping block coded data interleaved according to a modulation map to generate modulated input data; converting the modulated input data to the frequency domain by an inverse fast Fourier transform (IFFT); and mixing the modulated input data converted to the frequency domain with a carrier signal to generate output data. A wireless transmitter and / or wireless receiver may be referred to herein as a wireless transceiver.
[0053] Examples described herein include systems and methods for training computing devices with coefficient data. In some examples, a wireless transmitter may receive input associated with an RF wireless transmission. The wireless transmitter may perform operations as an RF front end, such as modulating input data for RF wireless transmission. Output data generated by the wireless transmitter may be compared with the input data to generate coefficient data. A computing device that receives and compares the output data and other corresponding input data and corresponding subsequent output data may be trained to generate coefficient data based on the operation of a specially designed wireless transmitter, so that mixing any input data using the coefficient data generates an approximation of the output data as if it were processed by a specially designed wireless transmitter. The coefficient data may also be stored in a coefficient database, where each set of coefficient data corresponds to a wireless protocol that can be used for data transmission in the RF domain.
[0054] In some examples, a computing device may receive a processing mode selection, such as from a user interacting with the computing system. The processing mode selection may indicate a particular processing mode of the computing system. As further described below, the processing mode may correspond to a single processing mode, a multiple processing mode, or a full processing mode. As examples, a full processing mode may be a processing mode representing a wireless transmitter (e.g., a wireless transmitter processing mode), or a processing mode representing a wireless receiver (e.g., a wireless receiver processing mode). For example, a wireless transmitter mode may include operation of an RF front end. Thus, when the computing device receives such a processing mode selection, the computing device may provide output data representing data transmissions processed according to the wireless transmitter mode.
[0055] In general, any aspect of a wireless protocol can be used to generate coefficient data, which in turn can be used to mix input data to generate output data, which represents the aspect of the wireless protocol processed in hardware that implements the aspect of the wireless protocol. For example, an FPGA can be used to process IFFT to transmit various data transmissions according to a wireless protocol incorporating IFFT. As disclosed herein, a computing system that mixes input data with coefficient data specific to an IFFT operation can be used to generate output data representing an IFFT, as if the input data were processed in the above-mentioned FPGA configured to process IFFT. In some instances, the computing system is advantageously executed in a more general setting compared to an FPGA that processes IFFT. An FPGA that implements IFFT may be a pre-designed hardware unit that is optimized to execute under a specific setting of IFFT (e.g., a 256-point IFFT setting). Therefore, an FPGA designed for a 256-point IFFT is limited to performing optimally for a wireless protocol that specifies a 256-point IFFT. However, if a specific implementation of a wireless protocol specifies that a 512-point FFT will be performed by an FPGA, the FPGA may not perform optimally in the setting. Using examples of the systems and methods described herein, a computing system can advantageously be configured to operate as a 256-point IFFT or a 512-point IFFT, depending on system or user input (e.g., processing mode selection), thereby allowing the computing system to perform optimally in more settings than an FPGA configured to implement a particular type of IFFT.
[0056] Although some examples have been described herein with respect to IFFT, it should be appreciated that various aspects of the wireless protocol may be processed by mixing input data with coefficient data to generate output data representative of the aspects. For example, other aspects of the wireless protocol that may be implemented by coefficient data mixed with input data include, but are not limited to: baseband processing of a wireless transmitter, baseband processing of a wireless receiver, processing of a digital front-end transmitter (e.g., a digital RF transistor), analog-to-digital conversion (ADC) processing, digital-to-analog (DAC) conversion processing, digital up-conversion (DUC), digital down-conversion (DDC), direct digital synthesizer (DDS) processing, DDC with DC offset compensation, digital pre-distortion (DPD), peak-to-average power ratio (PAPR) determination, crest factor reduction (CFR) determination, pulse shaping, image suppression, delay / gain / imbalance compensation, noise shaping, numerically controlled oscillator (NCO), self-interference cancellation (SIC), any modulation algorithm, any error correction coding or decoding algorithm, channel estimation, any pre-coding algorithm, and combinations thereof.
[0057] A multi-layer neural network (NN) and / or a multi-layer recurrent neural network (RNN) may be used to transmit or receive wireless input data (e.g., as wireless input data to be transmitted via an antenna or noisy wireless input data received at an antenna). The NN and / or RNN may have nonlinear mapping and distributed processing capabilities, which may be beneficial to many wireless systems, such as those involving processing wireless input data with time-varying wireless channels (e.g., autonomous vehicle networks, drone networks, or Internet of Things (IoT) networks). In this way, the neural networks and / or recurrent neural networks described herein may be used to implement wireless systems associated with various wireless protocols (e.g., 5G wireless protocols).
[0058] Figure 1A 1 is a schematic diagram of an electronic device 110 arranged in a system 100 according to an example described herein. The electronic device 110 is coupled to a memory 140 via a network 120, or to a memory 145, each of which can store coefficient data. The coefficient data stored in the memory 140 or the memory 145 can include coefficient data, which can be mixed with input data received by the electronic device 110 in the example described herein. The electronic device 110 also includes a processing unit 112, which can interact with a mode configurable control 105 and a memory 107, both of which can be encoded with instructions that can be executed by the processing unit 112. In some implementations, the mode configurable control 105 can be implemented as a memory. As used herein, a memory can refer to a computer-readable medium, which can include both storage media and communication media. Example computer-readable media can include volatile and non-volatile, removable and non-removable media implemented in any method or technology to store information such as computer-readable instructions. The mode configurable control 105 includes executable instructions 115 for mixing input data with coefficients; and the memory 107 includes executable instructions 117 for implementing the wireless processing mode. For example, the mode configuration control 105 can be implemented using circuitry (e.g., logic), one or more processors, microcontrollers, controllers, or other elements. The mode configuration control 105 can select specific weights and / or other parameters (e.g., from the memory 145 storing weights) and provide those weights and / or other parameters to the wireless processing mode. Figures 2A to 2C In some examples, the weights and / or other parameters stored in memory 145 may be associated with a particular wireless processing mode. During operation, mode configuration control 105 may be used to select the weights and / or other parameters in memory 145 based on the indicated wireless processing mode, which may be supplied from memory 107 executing executable instructions 117 for implementing the wireless processing mode.
[0059] Processing unit 112 may be implemented using, for example, one or more processors having any number of cores. In some examples, processing unit 112 may include circuitry, including custom circuitry, and / or firmware for performing the functions described herein. For example, the circuitry may include a multiplication unit / accumulation unit for performing the functions described, as described herein. Processing unit 112 may be of any type including, but not limited to, a microprocessor or a digital signal processor (DSP), or any combination thereof. For example, processing unit 112 may include a hierarchy of caches (e.g., a first level cache and a second level cache), cores, and registers. An example processor core may include an arithmetic logic unit (ALU), a bit manipulation unit, a multiplication unit, an accumulation unit, an adder unit, a lookup table unit, a memory lookup unit, or any combination thereof. Reference herein, for example, Figure 2A , 2B Or 2C describes an instance of processing unit 112.
[0060] The mode configurable control 105 may be encoded, for example, by executable instructions 115 for mixing input data with coefficient data, for example to mix the input data with coefficient data at the processing unit 112. For example, in the context of a 5G wireless transmission system, the executable instructions 115 for mixing input data with coefficient data may include instructions for providing output data representing input data processed according to a wireless protocol for the 5G wireless transmission to the antennas 131, 133, 135. The executable instructions 115 for mixing input data with coefficient data may further include instructions for multiplying a portion of the input data with the coefficient data to generate a coefficient multiplication result, and accumulating the coefficient multiplication result to further multiply and accumulate with other input data and coefficient data, examples of which are described herein. For example, to generate the coefficient multiplication result, a first multiplication / accumulation processing unit (MAC unit) layer may calculate the input data with a plurality of coefficients to generate such a coefficient multiplication result, or a first processing result of the first MAC unit layer. Continuing with the example, to provide output data, the additional layers of MAC units may calculate the first processing result with the additional plurality of coefficients to generate the additional coefficient multiplication result, or the second processing result of the additional layers of MAC units. The MLU of the last layer of the additional layers of MAC units may provide output data based on the second processing result. Thus, the executable instructions 115 for mixing input data with coefficient data may include instructions for different types of hardware implementations, such as Figures 2A to 2C Various sets of executable instructions for those hardware implementations shown in to implement such mixing of input data and coefficient data.
[0061] The input data to be mixed may be used for transmission in the RF domain or may be obtained from the RF domain. The electronic device 110 may transmit or transceive such transmissions via a plurality of antennas coupled to the electronic device 110. Each antenna 131, 133, and 135 may be associated with a different RF frequency at which it transceives, transmits, or receives RF transmissions. Antennas 131, 133, and 135 are coupled to the electronic device 110 as receiving antennas. Such antennas may obtain RF transmissions, which may be provided to the processing unit 112 for processing as input data. For example, each antenna 131, 133, and 135 may represent input data obtained on a corresponding wireless channel, such as a wireless channel associated with a corresponding frequency of the corresponding antenna 131, 133, and 135. Therefore, each RF transmission obtained at the corresponding antenna 131, 133, and 135 may be referred to as a portion of the input data to be mixed. As transmitting antennas, antennas 131, 133, and 135 may provide output data based on a varying portion of the input data to be mixed. For example, each portion of the mixed input data may represent a respective wireless channel, with output data provided for transmission as a multiple-input multiple-output ("MIMO") RF signal at antennas 131, 133, and 135. In some implementations, the number of additional layers of MAC and MLU units is associated with the number of wireless channels.
[0062] For example, the memory 107 may be encoded with executable instructions 117 for implementing a wireless processing mode. The processing mode selection may cause the electronic device 110 to receive input data for transmission based on the processing mode selection. In general, the electronic device 110 may process the input data according to various processing modes. In an example, the multi-processing mode may include at least two aspects of the wireless protocol, while the single processing mode includes one aspect of the wireless protocol. Among other aspects, aspects of the wireless protocol may include: fast Fourier transform (FFT) processing, inverse fast Fourier transform (IFFT) processing, decoding, turbo decoding, Reed Solomon processing, decoding, interleaving, deinterleaving, modulation mapping, demodulation mapping, scrambling, descrambling, or channel estimation. In some examples, the output data may be formatted so that the output data can be received by another wireless processing unit for further processing. For example, the computing system may operate as a decoding operation in a single processing mode to output decoded data. Alternatively, in a specific example, the computing system may operate as a turbo decoding operation in a single processing to output turbo decoded data. The decoded data may be formatted to be received by another wireless processing unit, such as an interleaving that may be processed differently by a computing system or another computing system (e.g., a cloud computing system). The processing mode selection may be received via a user interface 114. In some instances, the processing mode selection may be received by decoding and / or inspecting a portion of the incoming input data. For example, the electronic device 110 may identify that the input data is intended for processing using a particular processing mode, for example, by identifying a pattern or other signature in the input data indicating the processing mode. In some instances, the processing mode selection may also be received from the network 120. When receiving an incoming data stream from the network 120, the electronic device 110 may receive the processing mode selection in the data stream from a control channel or other control plane implemented at the network 120. For example, the electronic device 110 implemented as a user equipment (UE) may receive radio resource control (RRC) signaling from the network 120 via a control channel, wherein the RRC includes a configuration or flag for the processing mode selection to be implemented by the electronic device 110.
[0063] The user interface 114 may be implemented with any of a plurality of input devices, including but not limited to a touch screen, a keyboard, a mouse, a microphone, or a combination thereof. The user interface 114 may receive input from a user, such as regarding a processing mode selection, to specify a processing mode for the processing unit 112. The user interface 114 may transmit the user input to the mode configurable control 105 and the memory 107 for processing the user input. The example user interface 114 includes a serial interface controller or a parallel interface controller, which may be configured to communicate with an external input device (e.g., a keyboard, a mouse, a pen, a voice input device, a touch input device, etc.).
[0064] The network 120 may include a wired network or direct-wired connection, and wireless media such as acoustic, radio frequency (RF), microwave, infrared (IR) and other wireless media.
[0065] Memories 107, 140, and 145 (or mode configurable controls 105, if implemented as memory) may be implemented using any storage medium accessible by processing unit 112. For example, RAM, ROM, solid-state memory, flash memory, disk drives, system memory, optical storage devices, or combinations thereof may be used to implement mode configurable controls or memories 107, 140, and 145. For example, mode configurable controls 105 and memory 107 may store correlations between coefficients and wireless protocols and / or processing modes described herein.
[0066] The electronic device 110 may be implemented using any of a variety of computing systems, including but not limited to one or more desktop computers, servers, laptop computers, or other computers. The electronic device 110 typically includes one or more processing units 112. The computing system 100 may be implemented as a mobile communication device using any user communication device, including but not limited to a desktop computer, a laptop computer, a cellular phone, a tablet computer, an appliance, an automobile, or a combination thereof. The electronic device 110 may be programmed with a mobile application (e.g., a processing unit 112 and a computer-readable medium encoded with instructions that, when executed, cause the electronic device 110 to perform the described functions) to mix input data with coefficient data or to specify a processing mode. For example, the electronic device 110 may be programmed to receive an indication from a touch screen of a mobile communication device that multiple processing modes will be used for data received via 5G wireless data transmission.
[0067] It should be understood that the arrangement of the computing system of system 100 can be quite flexible, and although not shown, it should be understood that system 100 can include many electronic devices 110 that can be connected via network 120 and can operate in conjunction with each other to perform the systems and methods described herein. In some examples, memory 145 and / or storage 140 can be implemented using the same media, and in other examples, can be implemented using different media. For example, while memory 140 is in Figure 1A 105. In the example embodiment, the memory 140 is shown as being coupled to the network 120, but it should be understood that the memory 140 may also be implemented by the electronic device 110 as part of the mode configurable control 105. Figure 1A A single user interface 114 is shown in FIG. 1 , but it should be understood that the electronic device 110 may further include any number of input devices, output devices, and / or peripheral components. For example, the user interface 114 may be an interface of a mobile communication device.
[0068] Figure 1Bis a schematic diagram of a neural network arranged according to the examples described herein. Neural network 150 includes three levels (eg, layers). Figure 1B 16, but in other examples, any number of stages may be used. A first level of neural network 150 includes node 168, node 169, node 172, and node 174. A second level of neural network 150 includes combiner 152, combiner 154, combiner 156, and combiner 158. A third level of neural network 150 includes combiner 160, combiner 162, combiner 164, and combiner 166. In other examples, additional, fewer, and / or different components may be used.
[0069] In general, a neural network comprising multiple levels of nodes may be used. A node may be implemented using a processing unit (e.g., processing unit 112) that may perform one or more functions on input received from a previous level and provide output of the functions to the next level of the neural network. A processing unit may be implemented using, for example, one or more processors, controllers, and / or custom circuitry, such as an application specific integrated circuit (ASIC) and / or a field programmable gate array (FPGA). In some examples, a processing unit may use a processor or processors associated with the processing unit. Figures 2A to 2C The neural network 150 may be implemented by any combination of one or more processing units described herein. The processing units may be implemented as combiners and / or summers and / or any other structure for performing functions assigned to the processing units. In some examples, certain elements of the neural network described herein perform weighted sums, for example, which may be implemented using one or more multiplication / accumulation units, which may be implemented using a processor and / or other circuitry. In an example, the neural network 150 may be implemented by the electronic device 110.
[0070] exist Figure 1B In the example of , neural network 150 may have an input layer, which may be a first level of a neural network including nodes 168, 169, 172, and 174. Nodes 168, 169, 172, and 174 may implement linear functions that may provide input signals (e.g., x1(n), x2(n), ... xm(n)) to another level (e.g., a 'hidden level' or 'hidden layer') of the neural network. Thus, in Figure 1BIn an example of , m signals of input data may be provided to an input stage (e.g., an input layer) of a neural network during operation. In some examples, the input data may be wireless input data associated with a wireless protocol (e.g., input data to be transmitted as wireless signals in transmission according to the wireless protocol, and / or data received according to the wireless protocol). In some implementations, the input data may also be received wireless signals to be processed (e.g., decoded) at the neural network 150, e.g., because the neural network 150 may receive and process received wireless signals as a wireless receiver configured to operate according to a particular wireless protocol. The m signals of input data may be output by a first stage of the neural network 150 to a next stage of the neural network 150. In some examples, the connection between the first stage and the second stage of the neural network 150 may not be weighted, e.g., in some examples, a processing unit in the second stage may receive the signal unchanged from the first stage. Each of the input signals may be provided to a plurality of processing units in the processing unit in the next stage. Although an input layer is shown, in some examples, the input layer may not be present.
[0071] Neural network 150 may have a next layer, which in some examples may be referred to as a 'hidden layer'. The next layer may include combiner 152, combiner 154, combiner 156, and combiner 158, although any number of elements may be used. Although the processing elements in the second stage of neural network 150 are referred to as combiners, typically the processing elements in the second stage may perform a non-linear activation function using the input signal received at the processing element. Any number of non-linear activation functions may be used. Examples of functions that may be used include Gaussian functions, such as Examples of functions that can be used include polyquadratic functions, such as f(r)=(r 2 +σ 2 ) 1 / 2 Examples of functions that can be used include inverse polyquadratic functions, such as f(r)=(r 2 +σ 2 ) -1 / 2 Examples of functions that can be used include thin plate spline functions, such as f(r) = r 2 log(r). Examples of functions that can be used include piecewise linear functions, such as Examples of functions that can be used include cubic approximation functions, such as In these example functions, σ represents a true parameter (eg, a scaling parameter), and r is the distance between the input vector and the current vector. The distance can be measured using any of a variety of metrics, including the Euclidean norm.
[0072] Each element in the 'hidden layer' may receive as input a selected signal (e.g., some or all) of the input data. For example, each element in the 'hidden layer' may receive as input the output of a plurality of selected units (e.g., some or all of the units) in the input layer. For example, combiner 152 may receive as input the output of node 168, node 169, node 172, and node 174. Although in Figure 1B A single 'hidden layer' is shown by way of example in , but any number of 'hidden layers' may be present and may be connected in series. Although four units are shown in the 'hidden layer', any number may be used, and their number may be the same or different from the number of nodes in the input layer and / or the number of nodes in any other hidden layer. The nodes in the hidden layer may evaluate at least one nonlinear function using a combination of data received at the hidden layer nodes (e.g., elements). In this way, the hidden layer may provide intermediate data at the output of one or more hidden layers.
[0073] Neural network 150 may have an output layer. Figure 1B The output layer in the example of may include combiner 160, combiner 162, combiner 164 and combiner 166, but any number of elements may be used. Although the processing elements in the output stage of neural network 150 are referred to as combiners, the processing elements in the output may generally use the intermediate processing results from the last 'hidden layer' in the neural network to perform any combination or other operations. Each element in the output layer may receive the selected intermediate processing results (e.g., some or all) of the data provided by the last 'hidden layer' as input. For example, combiner 160 may receive the outputs of combiner 152, combiner 154, combiner 156 and combiner 158 as input. The connection between the hidden layer and the output layer may be weighted. For example, a set of weights W may be specified. In some examples, there is one weight for each connection between a hidden layer node and an output layer node. In some examples, for each hidden layer node, there may be one weight that may be applied to the data provided by the node to each connected output node. Other distributions of weights may also be used. The weights may be multiplied by the output of the hidden layer before providing the output to the output layer. In this way, the output layer may perform a summation of the weighted inputs. Thus, the output of the neural network 150 (e.g., the output of the output layer) may be referred to as a weighted sum. The output layer may thus use the weights to combine intermediate data received from one or more hidden layers to provide output data.
[0074] In some examples, neural network 150 may be configured to provide L output signals representing processed data corresponding to m input signals. Figure 1B In the example, the m input signals are shown as (x 1 (n),x 2 (n),...xm (n)), and provides L output signals (y 1 (n),y 2 (n),...y L (n)). The neural network 150 may be trained such that the weights W used by the elements of the hidden layer and / or the functions provided cause the neural network 150 to provide an output signal representing processed data corresponding to the m wireless input signals. The input signals may be wireless input data associated with a wireless protocol (e.g., input data to be transmitted as wireless signals in transmission according to the wireless protocol), and the weights and / or functions provided by the elements of the hidden layer may be selected according to the wireless protocol. The input data may also be received wireless signals to be processed (e.g., decoded) at the neural network 150, e.g., because the neural network 150 may receive and process received wireless signals as a wireless receiver configured to operate according to a particular wireless protocol. Thus, when the neural network 150 is used to transmit and receive, the neural network 150 may be trained multiple times, e.g., at least twice for each wireless protocol that may be used to provide wireless input data to the neural network 150, e.g., once for transmitting wireless input data as a wireless transmitter configured to operate according to the particular wireless protocol, and once for receiving wireless input data as a wireless receiver configured to operate according to the particular wireless protocol.
[0075] Examples of trainable neural networks. Training generally refers to the process of determining weights, functions, and / or other properties to be utilized by a neural network to produce a desired transformation of input data to output data. In some examples, the neural network described herein may be trained to transform wireless input data to output data (e.g., input data to be transmitted in a transmission according to a wireless protocol). In some examples, the neural network described herein may be trained to decode and / or process received wireless input data into output data (e.g., an estimate of received wireless data as it was received and decoded by a wireless receiver based on data from the wireless receiver on which the neural network 150 is trained). In some implementations, the neural network may be used to reduce noise present in the input data. For example, some received wireless input data may be received via a noisy wireless channel, which may introduce noise into the received wireless signal in both the time domain and the frequency domain. Using a time delayed version of the wireless input data in the RNN may compensate for the noise introduced in the time domain and the frequency domain because the RNN utilizes corresponding time and frequency correlations associated with the time delayed version of the wireless input data. In this way, the neural network may be used to reduce and / or improve errors that may be introduced by noise.
[0076] Training as described herein may be supervised or unsupervised in various examples. In some examples, training may be performed using known pairs of expected input and desired output data. For example, training may utilize known pairs of wireless input data and output data to train a neural network to receive and process subsequent wireless input data into output data (e.g., input data to be transmitted in a transmission according to a wireless protocol). In some examples, training may utilize known noisy wireless data and output data pairs to train a neural network to decode subsequent noisy wireless input data into output data (e.g., input data received from a noisy wireless channel according to a wireless protocol). In some examples, training may utilize known pairs of noisy input data and output data to train a neural network to provide output data with reduced noise compared to the input noisy wireless data. Examples of training may include determining a signal to be transmitted by, for example, Figure 1B In some examples, the same neural network hardware is used during training as will be used during operation. However, in some examples, different neural network hardware may be used during training, and the weights, functions, or other properties determined during training may be stored for use by other neural network hardware during operation. For example, the weights, functions, or other properties determined during training may be stored in the neural network hardware of the electronic device 110. Figure 1A and / or Figures 2A to 2C The hardware implementation described in utilises a processing unit 112 (e.g. Figures 2A to 2C any combination of one or more processing units shown in ).
[0077] An example of training can be described mathematically. For example, consider the input data at time (n), given by: X(n) = [x 1 (n),x 2 (n),…x m (n)] T The center vector of each element in the hidden layer of neural network 150 (eg, combiner 152, combiner 154, combiner 156, and combiner 158) may be represented as C i (for i=1,2,…,H, where H is the number of elements in the hidden layer).
[0078] The output of each element in the hidden layer can then be given as:
[0079] h i (n) = f i (‖X(n)-C i ‖)For (i=1,2,……,H) (1)
[0080] The connection between the last hidden layer and the output layer can be weighted. Each element in the output layer can have a linear input-output relationship so that it can perform a summation (e.g., a weighted summation). Therefore, the output of the i-th element in the output layer at time n can be written as:
[0081]
[0082] For (i=1,2,…,L), and where L is the number of elements in the output layer, and W ij is the connection weight between the jth element in the hidden layer and the ith element in the output layer.
[0083] In general, neural network architectures (e.g. Figure 1B The neural network 150 may include multiple elements and may have center vectors distributed in the input domain, so that the neural network can approximate a nonlinear multidimensional function, and thus can approximate a wireless receiver or transmitter configured to receive or transmit (or transceive) input data (e.g., a signal as a wireless signal or a received wireless signal) as output data (e.g., a wireless transmission signal or a decoded signal from a received wireless signal). In general, the selection of the transfer function used by the elements in the hidden layer may not affect the mapping performance of the neural network, and thus in some instances, functions that can be conveniently implemented in hardware may be used. For example, thin plate spline functions and / or Gaussian functions may be used in various instances and may provide sufficient approximation capabilities. Other functions may also be used.
[0084] An instance of a neural network may thus be specified by attributes (e.g., parameters). In some instances, two sets of parameters may be used to specify a neural network: connection weights and a center vector (e.g., threshold). The parameters may be determined from selected input data (e.g., encoded input data) by solving an optimization function. An example optimization function may be given as:
[0085]
[0086] where M is the number of trained input vectors (e.g., trained encoded data inputs) and Y(n) is the output vector calculated from the sample input vector using equations (1) and (2) above, and is the corresponding desired (eg, known) output vector. The output vector Y(n) can be written as: Y(n) = [y 1 (n),y 2 (n),………y L (n)] T
[0087] Various methods (e.g., gradient descent procedures) can be used to solve the optimization function. However, in some instances, another method can be used to determine the parameters of the neural network, which can generally include two steps: (1) determining the center vector C i (i=1,2,…,H), and (2) determine the weights.
[0088] In some instances, the center vector may be selected from a subset of available sample vectors. In such instances, the number of elements in the hidden layer may be relatively large to cover the entire input domain. Therefore, in some instances, it may be necessary to apply the k-means clustering algorithm. Typically, the k-means clustering algorithm distributes the center vector according to the natural measurement of the attractor (for example, if the density of data points is high, the density of the center is high). The k-means clustering algorithm can find a set of cluster centers and partition the training samples into subsets. Each cluster center can be associated with one of the H hidden layer elements in this network. The data can be partitioned in such a way that the training points are assigned to the cluster with the nearest center. The cluster center corresponds to one of the minimum values of the optimization function. The example optimization function used with the k-means clustering algorithm can be given as:
[0089]
[0090] Among them B jn The cluster partitions or membership functions that form an HxM matrix. Each column may represent an available sample vector (eg, known input data), and each row may represent a cluster. Each column may contain a single '1' in the row corresponding to the cluster closest to the training point, and zeros elsewhere.
[0091] The center of each cluster can be initialized to a different randomly selected training point. Each training instance can then be assigned to the element closest to it. When all training points have been assigned, the average position of the training points for each cluster can be found and the cluster center moved to that point. The clusters can become the desired centers of the hidden layer elements.
[0092] In some examples, for some transfer functions (e.g., Gaussian functions), a scaling factor σ may be determined, and the scaling factor may be determined before determining the connection weights. The scaling factor may be selected to cover the training points to allow for a smooth fit of the desired network output. Typically, this means that any point within the convex hull of the processing element center may significantly activate more than one element. To achieve this goal, each hidden layer element may activate at least one other hidden layer element to a significant degree. A suitable method for determining the scaling parameter σ may be based on the P nearest neighbor heuristic, which may be given as,
[0093]
[0094] Among them C j(for i = 1, 2, ..., H) is C i The P nearest neighbors.
[0095] The connection weights may additionally or alternatively be determined during training. In the example of a neural network with one hidden layer of weighted connections and an output element that is a summing unit, for example Figure 1B Once the center vector and scaling parameters of the neural network 150 have been determined, the optimization function of equation (3) can be transformed into a linear least squares problem. The linear least squares problem can be written as
[0096]
[0097] Where W = {Wij} is the LxH matrix of connection weights, F is the HxM matrix of the outputs of the hidden layer processing elements, and its matrix elements use F in =f i (‖X(n)-C i ‖)(i=1,2,……,H;n=1,2,……,M)calculate, and is the LxM matrix of the desired (eg, known) output. The connection weight matrix W can be found from equation (5) and can be written as follows:
[0098]
[0099] where F + is the pseudo-inverse of F. In this way, the above may provide a batch method for determining the connection weights of a neural network. It may be applied, for example, to situations where all input sample sets may be available simultaneously. In some examples, each new sample set may become available recursively, such as in a recursive least squares algorithm (RLS). In such cases, the connection weights may be determined as follows.
[0100] First, the connection weights can be initialized to any value (for example, random values can be used). The output vector Y(n) can be calculated using equation (2). The error term e for each output element in the output layer can be calculated as follows i (n):
[0101]
[0102] The connection weights may then be adjusted based on the error term, for example as follows:
[0103] W ij (n+1)=W ij (n)+γe i (n)f j (‖X(n)-C i ‖) (7)(i=1,2,…,L; j=1,2,…,M)
[0104] where γ is a learning rate parameter that can be fixed or time-varying.
[0105] The total error can be calculated based on the output from the output layer and the desired (known) data:
[0106]
[0107] The process can be iterated by again calculating the new output vector, the error term, and again adjusting the connection weights. The process can continue until a weight is identified that reduces the error to equal to or less than a threshold error.
[0108] therefore, Figure 1B The neural network 150 of the embodiment may be trained to determine parameters (e.g., weights) for the neural network 150 to use to perform a particular mapping between input data and output data. For example, training the neural network 150 may provide a set of parameters for use when receiving wireless input data, thereby decoding the data into output data as a wireless receiver would. Additionally or alternatively, training the neural network 150 may provide the set of parameters for use when receiving wireless input data, as experienced in a noisy wireless channel. For example, the neural network 150 (and / or another neural network) may be trained multiple times using different known input / output data pairs. For example, in the context of decoding a received wireless signal, known input / output pairs for different demodulation mappings may be used. In some examples, in the same context of decoding a received wireless signal, known input / output pairs for more than one aspect of a wireless receiver processing mode may be utilized, such as a decoding aspect, a deinterleaving aspect, and a demodulation aspect, or any combination of wireless receiver processing aspects. As another example, training neural network 150 may provide a parameter, , for use in transforming input data into output data to be transmitted as a wireless signal, as a wireless transmitter will process such input data.
[0109] Multiple trainings may produce multiple sets of connection weights. For example, a different set of weights may be determined for each of a plurality of wireless protocols - for example, one set of weights may be determined for transmitting wireless input data according to a wireless protocol, and another set of weights may be determined for receiving and decoding wireless input data according to the wireless protocol.
[0110] Recall that the examples provided are only Figure 1BThe structure of the neural network 150. Other multi-layer neural network structures can be used in other examples. In addition, the training process described herein is also provided by way of example. For example, other training techniques (e.g., learning algorithms) can be used to solve the local minimum problem and / or the vanishing gradient problem. The determined weights and / or vectors of each decoder can be obtained through an offline learning mode of the neural network, which can advantageously provide more resources and data. In some embodiments, for example, as shown in reference Figures 2A to 2C As described, the hidden layers of combiners (e.g., combiner 152, combiner 154, combiner 156, and combiner 158) may include multiply / accumulate (MAC) units, with each layer having additional MAC units. Such implementations that have accumulated intermediate processing results in respective processing elements (e.g., respective MAC units) may also include a memory lookup (MLU) unit configured to retrieve a plurality of coefficients and provide the plurality of coefficients as connection weights (e.g., determined weights) of the respective MAC unit layer to be mixed with input data.
[0111] Figure 1C is a schematic diagram of a recursive neural network arranged according to the examples described herein. Neural network 170 includes three stages (eg, layers): input node 171; combiner stages 173 and 175, and output node 177. Although in Figure 1C Three stages are shown in FIG. 1 , but in other examples, any number of stages may be used, for example, as shown in FIG. Figures 2A to 2C In some embodiments, the neural network 170 may have multiple combiner stages such that the output from one combiner stage is provided to another combiner stage until it is provided to the output node 177. Figure 2A As described, for example, there may be multiple combiner stages in the neural network 170. Figure 1C , delay units 175a, 175b, and 175c may be optional components of neural network 170. When such delay units 175a, 175b, and 175c are utilized as described herein, neural network 170 may be referred to as a recurrent neural network.
[0112] The first level of neural network 170 includes input node 171. Input node 171 can receive input data at various inputs of the recurrent neural network. In some examples, input node 171 can include multiple input nodes, such as Figure 1BThe second stage of the neural network 170 is a combiner stage, which includes combiner units 173a, 173b, 173c; and delay units 175a, 175b, 175c. Therefore, combiner unit 173 and delay unit 175 may be collectively referred to as a combiner stage. In some implementations, combiner units 173a, 173b, and 173c may correspond, for example, to Figure 1B The combiner 152, the combiner 154 and the combiner 156. Therefore, as for Figure 1B As described, where processing unit 112 implements such a combiner, processing unit 112 that typically implements combiner units 173a-c and delay units 175a-c in the second stage may use input data from input node 171 (e.g., input signal X 1 (n), X 2 (n) and X 3 (n)) to perform a nonlinear activation function. The third level of neural network 170 includes output node 177. In some examples, output node 177 may include Figure 1B 160, 162, and 164. Thus, in some examples, output node 177 may be referred to as a combiner stage. In other examples, additional, fewer, and / or different components may be used.
[0113] The recurrent neural network 170 includes delay units 175a, 175b, and 175c, which generate delayed versions of the outputs from the respective combiner units 173a-c based on receiving such output data from the respective combiner units 173a-c. In an example, the output data of the combiner units 173a-c may be represented as h(n); and, therefore, each of the delay units 175a-c delays the output data of the combiner units 173a-c to generate a delayed version of the output data from the combiner units 173a-c, which may be represented as h(nt). In various embodiments, the delay amount t may also be different, for example, one clock cycle, two clock cycles, or one hundred clock cycles. That is, the delay unit 175 may receive a clock signal and utilize the clock signal to identify the delay amount. In Figure 1C In the example of , the delayed version is delayed by one time period, where '1' represents a time period. The time period may correspond to any number of time units, such as a time period defined by a clock signal or a time period defined by another element of neural network 170.
[0114] exist Figure 1CContinuing in the example of , each delay unit 175a-c provides a delayed version of the output data from the combiner unit 173a-c as an input to the combiner unit 173a-c, optionally to operate as a recursive neural network. Such delay units 175a-c may provide corresponding delayed versions of the output data from the nodes of the combiner units 173a-c to the corresponding input units / nodes of the combiner units 173a-c. While utilizing the delayed versions of the output data from the combiner units 173a-c, the recursive neural network 170 may train weights at the combiner units 173a-c that incorporate time-varying aspects of the input data to be processed by such recursive neural network 170. In some examples, once trained, the input node 171 receives wireless input data to be received and processed in the recursive neural network 170 as a wireless receiver associated with a wireless protocol. Each input data stream may correspond to a different received signal or to be transmitted at a corresponding antenna (e.g., Figure 1A 131, 133, 135). Therefore, because the RNN 170 incorporates delayed versions of the output data from the combiner units 173a-c, the time-varying nature of the input data can provide faster and more efficient processing of the input data. For example, in the context of the input data being a received wireless signal, the time correlation of the received signal can be incorporated into the estimate of the decoded output data, so that the decoded processed data is a better estimate of the output data of the neural network than that not utilizing the delay units 175a-c. In some embodiments, in the context of receiving wireless signals from a noisy wireless channel, 170 provides individualized frequency band, time-correlated data for receiving and processing noisy wireless input data. Therefore, the recurrent neural network 170 can operate more efficiently in wireless networks where signals experience time variations in time-varying wireless channels, thereby partially causing noise in the received wireless signals. Although described in the context of r input data to be received and decoded, it will be appreciated that the recurrent neural network 170 may also be trained and applied to input data to be transmitted, such as where the recurrent neural network 170 is implemented as a wireless transmitter according to a wireless protocol.
[0115] In general, a recurrent neural network may include multiple levels of nodes. A node may be implemented using a processing unit (e.g., processing unit 112) that may perform one or more functions on inputs received from a previous level and provide outputs of the functions to the next level of the recurrent neural network. A processing unit may be implemented using, for example, one or more processors, controllers, and / or custom circuitry, such as an application specific integrated circuit (ASIC) and / or a field programmable gate array (FPGA). In some examples, a processing unit may use a processor or a processor associated with the node. Figures 2A to 2CThe neural network 170 may be implemented by any combination of one or more processing units 112 described herein. The processing units may be implemented as combiners and / or summers and / or any other structure for performing functions assigned to the processing units. In some examples, certain elements of the neural network described herein perform weighted sums, for example, may be implemented using one or more multiplication / accumulation units, which may be implemented using a processor and / or other circuitry. In an example, the neural network 170 may be implemented by the electronic device 110 using information about Figures 2A to 2C The invention may be implemented by any combination of one or more of the processing units described.
[0116] An example of recurrent neural network training and inference can be described mathematically. As an example, consider again the input data at time (n), given by: x(n) = [x 1 (n),x 2 (n),…x m (n)] T For each element in the hidden layer of the recurrent neural network 170 (e.g., including Figure 1B The center vector of the combiner unit 173 of combiner 152, combiner 154, combiner 156 and combiner 158 can be expressed as C i (for i=1,2,…,H, where H is the number of elements in the hidden layer).
[0117] The output of each element in the hidden layer can then be given as:
[0118] h i (n) = f i (‖X(n)+h i (nt)-C i ‖) for (i=1,2,……,H) (9)
[0119] t may be the delay at delay unit 175, such that the output of combiner unit 173 includes a delayed version of the output of combiner unit 173. In some examples, this may be referred to as feedback for combiner unit 173. Thus, each of the connections between the last hidden layer and the output layer may be weighted. Each element in the output layer may have a linear input-output relationship such that it may perform a summation (e.g., a weighted summation). Thus, the output of the i-th element in the output layer at time n may be written as:
[0120]
[0121] For (i=1,2,…,L), and where L is the number of elements in the output layer, and W ij is the connection weight between the jth element in the hidden layer and the ith element in the output layer.
[0122] Additionally or alternatively, although the description has been made with respect to a single stage (eg, second stage) of a combiner including combiner units 173a-c and delay units 175a-c, Figure 1C , but it will be appreciated that multiple stages of similar combiner stages may be included in the neural network 170 having different types of combiner units and different types of delay units having different delays, for example, as will now be referenced Figures 2A to 2C described.
[0123] Figure 2A is a schematic diagram of a processing unit 112 arranged in a system 200 according to examples described herein. Such a hardware implementation (e.g., system 200) may be used, for example, to implement one or more neural networks, such as Figure 1B Neural Network 150 or Figure 1C In addition or alternatively, in some implementations, the processing unit 112 may receive input data 210a, 210b, and 210c from such a computing system. The input data 210a, 210b, and 210c may be data obtained from a sensor or data stored in the memory 145. In some examples, the data stored in the memory 145 may be input data obtained from multiple antennas coupled to the electronic device 110 in which the processing unit 112 is implemented. In an example where the electronic device 110 is coupled to multiple antennas 131, 133, and 135, the input data 210a, 210b, and 210c may be data obtained from a sensor or data stored in the memory 145. In some examples, the data stored in the memory 145 may be input data obtained from multiple antennas coupled to the electronic device 110 in which the processing unit 112 is implemented. 1 (i, i-1) may correspond to a first RF transmission obtained at antenna 131 at a first frequency; input data 210bX 2 (i, i-1) may correspond to a second RF transmission obtained at antenna 133 at a second frequency; and input data 210c X m (i, i-1) may correspond to the mth RF transmission obtained at the mth frequency at the antenna 135. m may represent the number of antennas, each receiving a portion of the input data.
[0124] In some examples, m may also correspond to the number of wireless channels through which the input data is obtained; for example, in MIMO transmission, RF transmission may be obtained through multiple wireless channels at multiple antennas 131, 133, and 135. In examples where the input data is data to be transmitted (as opposed to obtained), the input data 210a, 210b, 210c may correspond to portions of the input data to be processed as RF transmission for multiple antennas. For example, the output data 230B(n) may be MIMO output signals to be transmitted at the antennas 131, 133, and 135 at an electronic device implementing the processing unit 112 of the computing system 200.
[0125] As shown in the representation of the input data signal, the input data 210a X1 (i, i-1) includes the current portion of the input data at time i and the previous portion of the input data at time i-1. For example, the current portion of the input data may be a sample obtained at antenna 131 at a particular time period (e.g., at time i), and the previous portion of the input data may be a sample obtained at antenna 131 at a time period before the particular time period (e.g., at time i-1). Therefore, the previous portion of the input data may be referred to as a time-delayed version of the current portion of the input data. For example, the portion of the input data at each time period may be obtained in a vector or matrix format. In an example, the current portion of the input data at time i may be a single value; and the previous portion of the input data at time i-1 may be a single value. Therefore, the input data 210aX 1 (i, i-1) may be a vector. In some examples, the current portion of the input data at time i may be a vector value; and the previous portion of the input data at time i-1 may be a vector value. Thus, the input data 210a X 1 (i,i-1) can be a matrix.
[0126] This input data obtained using the current portion and the previous portion of the input data may represent a Markov process, such that at least the causal relationship between the current sample and the previous sample may improve the accuracy of the weight estimation used to train the coefficient data to be utilized by the MAC unit and the MLU of the processing unit 112. As previously described, the input data 201 may represent data obtained at a first frequency and / or data obtained at a first wireless channel. Thus, the input data 210bb x 2 (i, i-1) may represent data obtained at a second frequency or at a second wireless channel, including a current portion of input data at time i and a previous portion of input data at time i-1. Also, in some examples, the number of input signals obtained by the processing unit 112 may be equal to the number of antennas coupled to the electronic device 110 implementing the processing unit 112. Thus, the input data 210c X m (i, i-1) may represent data obtained at the mth frequency or at the mth wireless channel, including a current portion of input data at time i and a previous portion of input data at time i-1.
[0127] Processing unit 112 may include multiply unit / accumulate (MAC) units 212a-c, 216a-b, and 220; delay units 213a-c, 217a-b, and 221; and memory lookup units (MLUs) 214a-c, 218a-b, and 222, which when mixed with input data obtained from memory 145 may generate output data (e.g., B(1)) 230. Each set of MAC units and MLU units having different element numbers may be referred to as a respective stage of a combiner for processing unit 112. For example, a first stage of a combiner includes MAC units 212a-c and MLUs 214a-c operating in conjunction with delay units 213a-c to form a first stage or "layer," as described with respect to the various combiner stages having "hidden" layers as the layers. Figure 1B Continuing in the example, the second stage of the combiner includes MAC units 216a-b and MLUs 218a-b operating in conjunction with delay units 217a-b to form a second stage or second layer of hidden layers. Also, the third stage of the combiner may be a single combiner including MAC unit 220 and MLU 222 operating in conjunction with delay unit 221 to form a third stage or third layer of hidden layers.
[0128] In an example of generating an RF transmission for transmission, the output data 230B(1) may be used as a MIMO RF signal to be transmitted at multiple antennas. In an example of obtaining an RF transmission obtained at multiple antennas, the output data 230B(1) may represent a demodulated, decoded signal transmitted by another RF electronic device. In any case, the processing unit 112 may provide instructions 115 stored in the mode configurable control 105 to cause the processing unit 112 to configure the multiplication units 212a-c, 216a-c, and 220 to multiply and / or accumulate delayed versions of the input data 210a, 210b, and 210c and the processing results from the delay units 213a-c, 217a-b, and 221 (e.g., the corresponding outputs of the corresponding layers of the MAC unit) with the coefficient data to generate the output data 230B(1). For example, the mode configurable control 105 may execute instructions that cause the memory 145 to provide weights and / or other parameters stored in the memory 145 that may be associated with a particular wireless processing mode to the MLUs 214a-c, 218a-b, and 222 as weights for the MAC units 212a-c, 216a-b, and 220 and the delay units 213a-c, 217a-b, and 221. During operation, the mode configuration control 105 may be used to select the weights and / or other parameters in the memory 145 based on an indicated wireless processing mode for the processing unit 112 (e.g., the indicated wireless processing mode is from the memory 107 executing executable instructions for the wireless processing mode 117).
[0129] As represented in the representation of the respective outputs of the respective layers of the MAC units (e.g., the outputs of MLUs 214a-c, 218a-b, and 222), the input data to each MAC unit 212a-c, 216a-b, and 220 includes a current portion of the input data at time i and a delayed version of the processing result at time i-1. For example, the current portion of the input data may be a sample obtained at antenna 131 at a particular time period (e.g., at time i), while the delayed version of the processing result may be obtained from the outputs of delay units 213a-c, 217a-b, and 221, which represent a time period prior to the particular time period (e.g., as a result of the introduced delay). Thus, when using this input data obtained from both the current period and at least one previous period, the output data 230B(1) may represent a Markov process such that the causal relationship between at least the data from the current time period and the previous time period may improve the accuracy of weight estimation for training coefficient data to be utilized by the MAC unit and MLU of the processing unit 112 or utilizing the processing unit 112 to make inferences about wireless input data or noisy wireless input data. As previously described, the input data 201 may represent data obtained at a first frequency and / or data obtained at a first wireless channel. Thus, the input data 210b X 2 (i, i-1) may represent data obtained at a second frequency or at a second wireless channel. Also, in some examples, the number of input signals obtained by the processing unit 112 may be equal to the number of antennas coupled to the electronic device 110 implementing the processing unit 112. Therefore, the input data 210c X m (i, i-1) may represent data obtained at the mth frequency or at the mth wireless channel, including the current portion of the input data at time i. Thus, in utilizing delayed versions of the output data from 213a-c, 217a-b, and 221, the recurrent neural network 170 provides personalized frequency band, time-dependent data for receiving and processing noisy wireless input data, such as received from a noisy wireless channel. Thus, the recurrent neural network 170 may operate more efficiently in wireless networks where signals experience time variations in time-varying wireless channels.
[0130] In an example of executing such an instruction 115 for mixing input data with coefficients, at the first level of the MAC units 212a-c and the MLUs 214a-c, the multiplication units / accumulation units 212a-c are configured to multiply and accumulate at least two operands from the corresponding input data 210a, 210b, or 210c and operands from the corresponding delay units 213a-c to generate a multiplication processing result provided to the MLUs 214a-c. For example, the multiplication units / accumulation units 212a-c may perform a multiplication-accumulation operation such that the three operands M, N, and T are multiplied and then added to P to generate a new version of P stored in its corresponding MLU 214a-c. Thus, the MLU 214a latches the multiplication processing result until the time to provide the stored multiplication processing result to the next level of the MAC units. MLUs 214a-c, 218a-b, and 222 may be implemented by any number of processing elements operating as memory lookup units, such as D, T, SR, and / or JK latches.
[0131] Figure 2A The MLUs 214a-c, 218a-b, and 222 shown in FIG. 2 may generally perform a predetermined nonlinear mapping from input to output. For example, the MLUs 214a-c, 218a-b, and 222 may be used to evaluate at least one nonlinear function. In some examples, the contents and sizes of the various MLUs 214a-c, 218a-b, and 222 depicted may be different and may be predetermined. In some examples, Figure 2A One or more of the MLUs 214a-c, 218a-b, and 222 shown in FIG. 2 may be replaced by a single merged MLU (e.g., table lookup). Examples of nonlinear mappings (e.g., functions) that may be performed by the MLUs 214a-c, 218a-b, and 222 include Gaussian functions, piecewise linear functions, sigmoid functions, thin plate spline functions, polyquadratic functions, cubic approximations, and inverse polyquadratic functions. Figure 1B Examples of Descriptive Functions. In some examples, selected MLUs 214a-c, 218a-b, and 222 may be bypassed and / or may be deactivated, which may allow the MLUs and their associated MAC units to be treated as unity gain elements.
[0132] In addition, in the example, MLU 214a provides the processing result to delay unit 213a. Delay unit 213a delays the processing result (eg, h 1 (i)) to produce a delayed version of the processing result (e.g., h 1 (i-1)) is output to the MAC unit 212a as the operand T. Figure 2AThe delay units 213a-c, 217a-b, and 221 of the first MAC unit layer are depicted as introducing a delay of '1', but it is understood that different amounts of delay may be introduced to the output of the first MAC unit layer. For example, introducing a sample delay of '1' (eg, h 1 (i-1)) may alternatively introduce a sample delay of '2', '4' or '100'. In various implementations, delay units 213a-c, 217a-b, and 221 may correspond to any number of processing units that may use a clock signal or other time-directed signal to introduce a delay into the processing circuitry, such as a flip-flop (e.g., a D flip-flop) and / or one or more various logic gates (e.g., AND, OR, NOR, etc. ...) that may operate as a delay unit.
[0133] In the example of a first hidden layer of a recurrent neural network, MLUs 214a-c may retrieve coefficient data stored in memory 145, which may be weights associated with weights to be applied to the first MAC unit layer to apply to both data from the current cycle and data from the previous cycle (e.g., delayed versions of the first layer processing results). For example, MLU 214a may be a table lookup to retrieve one or more coefficients to be applied to both operands M and N (e.g., specific coefficients associated with the first frequency) and additional coefficients to be applied to operand T. MLUs 214a-c also provide the generated multiplication processing results to the next layer of MAC units 216a-b and MLUs 218a-b. Additional layers of MAC units 216a, 216b and MAC unit 220 working in conjunction with MLUs 218a, 218b and MLU 222, respectively, may continue to process the multiplication results to generate output data 230B(1). Using such a circuitry arrangement, output data 230B(1) may be generated from input data 210a, 210b, and 210c.
[0134] Advantageously, for example, compared to Figure 2B 112, the processing units 112 of system 200 may utilize a reduced number of MAC units and / or MLUs. The number of MAC units and MLUs in each layer of processing units 112 is associated with the number of channels and / or the number of antennas coupled to the device in which the processing units 112 are implemented. For example, a first layer of MAC units and MLUs may include m number of those units, where m represents the number of antennas, each receiving a portion of the input data. Each subsequent layer may have a reduced portion of MAC units, delay units, and MLUs. As depicted, in Figure 2A, for example, when m=3, the second layer of MAC units 216a-b, delay units 217a-b, and MLUs 218a-b may include m-1 MAC units and MLUs. Thus, the last layer in processing unit 112 (including MAC unit 220, delay unit 221, and MLU 222) includes only one MAC, one delay unit, and one MLU. Because processing unit 112 utilizes input data 210a, 210b, and 210c that can represent a Markov process, the number of MAC units and MLUs in each subsequent layer of processing unit can be reduced without a significant loss in accuracy with respect to output data 230B(1); for example, when compared to processing unit 112 that includes the same number of MAC units and MLUs in each layer, a number similar to Figure 2B The number of processing units 112 in the system 250.
[0135] For example, the coefficient data from the memory 145 may be mixed with the input data 210a-210c and the delayed version of the processing result to generate the output data 230B(1). For example, the relationship between the coefficient data based on the input data 210a-c and the delayed version of the processing result and the output data 230B(1) may be expressed as:
[0136]
[0137] where a (m) 、a (m-1) 、a 1 are the coefficients for the first layer and outputs of the delay units 213a-c for the multiplication / accumulation units 212a-c, respectively; the second layer and outputs of the delay units 217a-b for the multiplication / accumulation units 216a-b; and the last layer and outputs of the delay unit 221 with the multiplication / accumulation unit 220; and where f(·) is a mapping relationship that can be performed by the memory lookup units 214a-c and 218a-b. As described above, the memory lookup units 214a-c and 218a-b retrieve coefficients to mix with the input data and the corresponding delayed versions of each layer of the MAC unit. Thus, the output data can be provided by manipulating the input data and the delayed versions of the MAC unit with the corresponding multiplication / accumulation units using the coefficient sets stored in the memory. The coefficient sets can be associated with the desired wireless protocol. The resulting mapped data can be manipulated by additional multiplication / accumulation units and additional delay units using additional coefficient sets associated with the desired wireless protocol stored in the memory. The set of coefficients multiplied at each stage of processing unit 112 may represent or provide an estimate of the processing of input data in specially designed hardware (eg, an FPGA).
[0138] Furthermore, it can be shown that the system 200 as represented by equation (11) can, in some instances, approximate any nonlinear mapping with arbitrarily small error, and that the mapping of the system 200 can be represented by the coefficients a (m) 、a (m-1) 、a 1 Determined. For example, if such coefficient data is specified, any mapping and processing between the input data 210a-210c and the output data 230 may be implemented by the system 200. For example, the coefficient data may represent a nonlinear mapping of the input data 210a-c to the output data B(1) 230. In some examples, the nonlinear mapping of the coefficient data may represent a Gaussian function, a piecewise linear function, a sigmoid function, a thin plate spline function, a multi-quadratic function, a cubic approximation, an inverse multi-quadratic function, or a combination thereof. In some examples, some or all of the memory lookup units 214a-c, 218a-b may be deactivated. For example, one or more of the memory lookup units 214a-c, 218a-b may operate as a gain unit with unity gain. This relationship, as derived from the circuit system arrangement depicted in the system 200, may be used to train the entities of the computing system 200 to generate coefficient data. For example, using equation (11), the entities of the computing system 200 may compare the input data with the output data to generate the coefficient data.
[0139] Each of the multiplication / accumulation units 212a-c, 216a-b, and 220 may include multiple multipliers, multiple accumulation units, and / or multiple adders. Any of the multiplication / accumulation units 212a-c, 216a-b, and 220 may be implemented using an ALU. In some examples, any of the multiplication / accumulation units 212a-c, 216a-b, and 220 may include one multiplier and one adder, each of which performs multiple multiplications and multiple additions, respectively. The input-output relationship of the multiplication / accumulation units 212a-c, 216a-b, and 220 may be expressed as:
[0140]
[0141] where "I" represents the number of multiplications performed in the unit, C i are coefficients accessible from a memory such as memory 145, and B in (i) represents the factors from the input data 210a-c or the outputs generated from the multiplication units / accumulation units 212a-c, 216a-b and 220. In the example, the outputs B of a group of multiplication units / accumulation units out Equal to the sum of coefficient data C i Multiply by the output B of another set of multiplication unit / accumulation unit in (i) B in(i) can also be input data, so that the output B of a set of multiplication units / accumulation units out Equal to the sum of coefficient data C i Multiplies the input data.
[0142] In some examples, instructions 115 are executed to determine if some coefficient data is the same. In this case, instructions 117 may be executed to facilitate selecting a single memory lookup unit for the same coefficients. For example, if the coefficient data to be retrieved by memory lookup units 214a and 214b are the same, then a single memory lookup unit 214 may replace memory lookup units 214a and 214b. Continuing in the example, instructions 117 may be further executed to configure memory lookup unit 214a to receive input from both multiplication unit / accumulation unit 212a and multiplication unit / accumulation unit 212b at different times or simultaneously.
[0143] In some examples where the processing unit 112 is used to provide input data to be transmitted in a transmission according to the wireless protocol, the output data B(1) 230 can be derived from the input of the system 200 in the following manner. The input data 210a-c can be represented as symbols to be modulated and generate the output data B(1) 230 of the DAC, thereby formatting the output data for transmission by an antenna (e.g., an RF antenna). In some examples, the input 210a-c can be expressed as:
[0144]
[0145] Where n is the time index, k is the subcarrier index, m is the symbol index, M is the number of symbols per subcarrier, K is the number of active subcarriers and N is the total number of subcarriers (e.g., the length of the discrete Fourier transform (DFT)), x(n) is the input data X(i,j) 210a-c, n] is the shaping filter coefficient, and d(k,m) is the decoded data associated with the mth symbol. In some examples where system 200 implements OFDM, equation (13) can be further generalized as:
[0146]
[0147] Where is the impulse response of the kth filter. Therefore, a filter with a rectangular impulse response can represent the input data X(i,j) 210a-c. And equation (15) can also be expressed as:
[0148]
[0149] Where B is the number of subbands, K b is the number of subcarriers in the bth subband, g k (b,n) is the impulse response of the corresponding kth filter in the bth subband.
[0150] Figure 2B is a schematic diagram of a processing unit 112 arranged in a system 250 according to examples described herein. Such a hardware implementation (e.g., system 250) may be used, for example, to implement one or more neural networks, such as Figure 1B Neural Network 150 or Figure 1C In addition or alternatively, in some embodiments, the processing unit 112 may be from Figure 1A The computing system 100 receives input data 210a, 210b and 210c. For example, Figure 1A In an example of , processor unit 112 may be implemented as processing unit 112 . Figure 2B The similarly described elements may be as described with respect to Figure 2A but may also include Figure 2B Additional features described. For example, Figure 2B Describing can be as about Figure 2A MAC units 262a-c and delay units 263a-c operate as described for MAC units 212a-c and delay units 213a-c. Therefore, their numerical indicators are relative to Figure 2A Offset 50 Figure 2B The system 250 including the processing element 112 also includes the elements of the processing element 112; for example, the MAC unit 266a operates similarly with respect to the MAC unit 216a. Figure 2A Additional features are highlighted in the processing element 112 of FIG. For example, Figure 2B The processing unit 112 further includes MAC units 266c and 270b-c; delay units 267c and 271b-c; and MLUs 268c and 272b-c, so that the output data is provided as 275a-c, rather than separately in Figure 2A 230. Advantageously, the system 250 including the processing element 112 may process the input data 260a-c to generate output data 275a-c having greater precision. For example, the output data 275a-c may be processed with additional coefficients retrieved at the MLU 268c and multiplied and / or accumulated by the additional MAC units 266c and 270b-c and the additional delay units 267c and 271b-c. For example, such additional processing may result in more accurate output data relative to providing output data in a processing mode representing a wireless transmitter (e.g., a wireless transmitter processing mode); a processing mode representing a wireless receiver (e.g., a wireless receiver processing mode); or any processing mode disclosed herein. In implementations where board space (e.g., a printed circuit board) is not a primary factor in the design, the output data 275a-c may be processed with additional coefficients that are retrieved at the MLU 268c and multiplied and / or accumulated by the additional MAC units 266c and 270b-c and the additional delay units 267c and 271b-c. For example, such additional processing may result in more accurate output data relative to providing output data in a processing mode representing a wireless transmitter (e.g., a wireless transmitter processing mode); a processing mode representing a wireless receiver (e.g., a wireless receiver processing mode); or any processing mode disclosed herein. Figure 2AAn implementation of the processing unit 112, Figure 2B An implementation of the processing unit 112 of Figure 2B The processing unit 112 has fewer components, so Figure 2A The processing unit 112 may occupy less board space.
[0151] Additionally or alternatively, the processing unit 112 of the system 250 may also be used for applications in which each portion of the output data 275a-c is to be transmitted as a MIMO signal on a corresponding antenna. For example, the output data 275a may be transmitted as part of a MIMO transmission at a first antenna having a first frequency; the output data 275b may be transmitted as a second portion of the MIMO transmission at a second antenna having a second frequency; and the output data 275c may be transmitted as an nth portion of the MIMO transmission at an nth antenna having an nth frequency.
[0152] Although in Figure 2A and 2B The processing element 112 is described as a single processing element 112 in the context of FIG. 1 , but features may also be implemented in Figure 1A The processing element 112 makes Figure 2A , 2B The description of a single processing element 112 in 2C or 2C may be interchangeably referred to as a processing element implemented in the examples described herein. Figure 2A , 2B or a single processing element 112 in 2C, but the electronic device 110 may have a similar Figure 1A The plurality of processing elements 112 may also have the following Figure 2A , 2B or the same features described for the single processing element 112 of 2C.
[0153] Figure 2C is a schematic diagram of a processing unit 112 arranged in a system 280 according to examples described herein. Such a hardware implementation (e.g., system 280) may be used, for example, to implement one or more neural networks, such as Figure 1B Neural Network 150 or Figure 1C Additionally or alternatively, in some embodiments, Figure 1A In the example of , processor unit 112 may be implemented as processing unit 112 . Figure 2C The similarly described elements may be as described with respect to Figure 2B operates as described, except that Figure 2B For example, Figure 2C Describing can be as about Figure 2BMAC units 282a-c and delay units 283a-c operate as described for MAC units 262a-c and delay units 263a-c. Therefore, their numerical indicators are relative to Figure 2B Offset 20 Figure 2C The elements of processing element 112 are similarly included; for example, MAC unit 286a operates similarly with respect to MAC unit 266a.
[0154] System 280 including processing element 112 also includes Figure 2B Additional features highlighted in the processing element 112 of FIG. Figure 2B , Figure 2C Delay units 283a, 283b and 283c are depicted. Thus, Figure 2C The processing unit 112 of the system 280 may include different arrangements of placement of inputs and outputs of delay units, as illustrated by delay units 283a, 283b, and 283c. For example, the output of MLU 288b may be provided to delay unit 283b to generate a delayed version of the processing result from the second layer of MAC units as input to the first MAC unit layer, e.g., as input to MAC unit 282b. Thus, the processing unit 112 of the system 280 illustrates that a delayed version of the processing result may be provided as input to other hidden layers, different from the processing unit 112 of the system 280. Figure 2B 250, which shows that the corresponding delayed versions are provided as input to the same layer where those delayed versions are generated (e.g., the output of MLU 268b is provided to delay unit 267b to generate a delayed version of MAC unit 266b in the same layer from which the processing results are output). Thus, in the example, the output B(n) 295c can even be provided from the last hidden layer to the first hidden layer (e.g., as input to MAC unit 282c).
[0155] Advantageously, for example, compared to Figure 2A Such delayed versions of the processing results of the processing unit 112 of the embodiment 2B may be provided as input to different or additional hidden layers to better compensate for the implementation Figure 2CThe invention also provides a method for modeling "higher order" memory effects in a recurrent neural network 170 of one or more processing units 112 of the present invention. For example, the high order memory effects model the effects of leading and lagging envelope signals used during training of the recurrent neural network 170 to provide output data representing an RF transmission processed according to a processing mode selection. The RF transmission may be input data to be transmitted in a transmission as a wireless signal according to a wireless protocol, data received according to a wireless protocol, and / or a received wireless signal to be processed (e.g., decoded). In an example, the recurrent neural network 170 that estimates the output data may include different delayed versions of the processing results corresponding to such leading and lagging envelopes (e.g., encapsulating various envelopes of the radio frequency transmission to be estimated). Thus, the processing unit 112 is implemented to incorporate such high order memory effects, such as for inference of the recurrent neural network 170 to provide output data 295a-c based on the input data 281a-c.
[0156] Additionally or alternatively, the processing unit 112 of the system 280 may also be used for applications in which each portion of the output data 295a-c is to be transmitted as a MIMO signal on a corresponding antenna. For example, the output data 295a may be transmitted as part of a MIMO transmission at a first antenna having a first frequency; the output data 295b may be transmitted as a second portion of the MIMO transmission at a second antenna having a second frequency; and the output data 295c may be transmitted as an nth portion of the MIMO transmission at an nth antenna having an nth frequency.
[0157] Although in Figure 2A , 2B 2C, processing element 112 is described as a single processing element 112, but features may also be implemented in Figure 1A The processing element 112 makes Figure 2A , 2B , 2C may be interchangeably referred to as processing elements implemented in the examples described herein. Thus, it can be appreciated that although reference is made to Figure 2A , 2B , 2C, but the electronic device 110 may have a single processing element 112 similar to Figure 1A The plurality of processing elements 112 may also have the following Figure 2A , 2B or the same features described for the single processing element 112 of 2C.
[0158] Figure 3 3 is a schematic diagram of a wireless transmitter 300. The wireless transmitter 300 receives a data signal 310 and performs operations to generate a wireless communication signal for transmission via an antenna 336. For example, the wireless transmitter 300 may be used to transmit a wireless signal to a wireless device 310. Figure 1A , 2A, 2B and 2C are implemented as wireless transmitters. Before transmitting the output data on the RF antenna 336, the transmitter outputs the data x N The (n) 310 is amplified by a power amplifier 332. The operations to the RF front end may be typically performed with analog circuitry or processed as digital baseband operations for implementing a digital front end. The operations of the RF front end include a scrambler 304, a decoder 308, an interleaver 312, a modulation map 316, a frame adaptation 320, an IFFT 324, a guard interval 326, and an up-conversion 328.
[0159] The scrambler 304 may convert the input data into a pseudo-random or random binary sequence. For example, the input data may be a transport layer source (such as an MPEG-2 transport stream and other data) that is converted into a pseudo-random binary sequence (PRBS) with a generator polynomial. Although described in the example of a generator polynomial, a variety of scramblers 304 are possible.
[0160] Decoder 308 can encode the data output from the scrambler to decode the data. For example, a Reed-Solomon (RS) encoder, a turbo encoder can be used as a first decoder to generate a parity block for each randomized transport packet fed by the scrambler 304. In some examples, the length of the parity block and the transport packet can vary according to various wireless protocols. Interleaver 312 can interleave the parity blocks output by decoder 308, for example, interleaver 312 can utilize convolutional byte interleaving. In some examples, additional decoding and interleaving can be performed after decoder 308 and interleaver 312. For example, additional decoding can include a second decoder, which can further decode the data output from the interleaver, for example, with an erasure type convolutional decoding with a specific constraint length. Additional interleaving can include an internal interleaver that forms a group of joined blocks. Although described in the context of RS decoding, turbo decoding, and erasure convolutional decoding, various decoders 308 are possible, such as a low density parity check (LDPC) decoder or a polar decoder. Although described in the context of convolutional byte interleaving, various interleavers 312 are possible.
[0161] The modulation map 316 may modulate the data output from the interleaver 312. For example, quadrature amplitude modulation (QAM) may be used to map the data by changing (e.g., modulating) the amplitude of the associated carrier. Various modulation maps may be used, including but not limited to: quadrature phase shift keying (QPSK), SCMA NOMA, and MUSA (multi-user shared access). The output from the modulation map 316 may be referred to as a data symbol. Although described in the context of QAM modulation, various modulation maps 316 are possible. Frame adaptation 320 may arrange the output from the modulation map according to a bit sequence representing the corresponding modulation symbol, carrier, and frame.
[0162] IFFT 324 may transform symbols that have been framed into subcarriers (e.g., by frame adaptation 320) into time domain symbols. Taking the example of a 5G wireless protocol solution, IFFT may be applied as an N-point IFFT:
[0163]
[0164] Where X n is the modulation symbol sent in the nth 5G subcarrier. Thus, the output of IFFT 324 may form a time domain 5G symbol. In some examples, IFFT 324 may be replaced by a pulse shaping filter or a polyphase filter bank to output symbols for up-conversion 328.
[0165] exist Figure 3 In the example of , guard interval 326 adds a guard interval to the time domain 5G symbol. For example, the guard interval can be a fractional length of the symbol duration added by repeating a portion of the end of the time domain 5G symbol at the beginning of the frame to reduce inter-symbol interference. For example, the guard interval can be a time period corresponding to the cyclic prefix portion of the 5G wireless protocol scheme.
[0166] The up-conversion 328 may up-convert the time domain 5G symbols to a specific radio frequency. For example, the time domain 5G symbols may be considered as a baseband frequency range, and the local oscillator may mix the frequency at which it oscillates with the 5G symbols to generate the 5G symbols at the oscillation frequency. A digital up-converter (DUC) may also be utilized to convert the time domain 5G symbols. Thus, the 5G symbols may be up-converted to a specific radio frequency for RF transmission.
[0167] Prior to transmission, at antenna 336, power amplifier 332 may amplify the transmitter output data x N (n) 310 to output data for RF transmission in the RF domain at antenna 336. Antenna 336 may be an antenna designed to radiate at a specific radio frequency. For example, antenna 336 may radiate at a frequency at which 5G symbols are up-converted. Thus, wireless transmitter 300 may transmit an RF transmission via antenna 336 based on data signal 310 received at scrambler 304. As described above with respect to Figure 3 As described, the operation of wireless transmitter 300 may include a variety of processing operations. Such operations may be implemented in a conventional wireless transmitter, where each operation is implemented by specially designed hardware for the respective operation. For example, a DSP processing unit may be specially designed to implement IFFT 324. As may be appreciated, additional operations of wireless transmitter 300 may be included in a conventional wireless receiver.
[0168] Figure 44 is a schematic diagram of a wireless receiver 400. The wireless receiver 400 receives input data X(i,j) 410 from an antenna 404 and performs wireless receiver operations to generate receiver output data at a descrambler 444. For example, the wireless receiver 400 may be used to convert Figure 1A , 2A , 2B and 2C are implemented as a wireless receiver. Antenna 404 may be an antenna designed to receive at a specific radio frequency. The operation of the wireless receiver may be performed by analog circuitry or processed as digital baseband operations for implementing a digital front end. The operation of the wireless receiver includes frequency down conversion 412, guard interval removal 416, fast Fourier transform 420, synchronization 424, channel estimation 428, demodulation mapping 432, deinterleaver 436, decoder 440 and descrambler 444.
[0169] Downconversion 412 may downconvert frequency domain symbols to a baseband processing range. For example, continuing in the example of a 5G implementation, frequency domain 5G symbols may be mixed with a local oscillator frequency to generate 5G symbols at a baseband frequency range. A digital downconverter (DUC) may also be utilized to convert frequency domain symbols. Thus, an RF transmission including time domain 5G symbols may be downconverted to baseband. Guard interval removal 416 may remove guard intervals from frequency domain 5G symbols. FFT 420 may transform time domain 5G symbols into frequency domain 5G symbols. Taking the example of a 5G wireless protocol solution, the FFT may be applied as an N-point FFT:
[0170]
[0171] Where X n is the modulation symbol sent in the nth 5G subcarrier. Thus, the output of FFT 420 may form a frequency domain 5G symbol. In some examples, FFT 420 may be replaced by a polyphase filter bank to output symbols for synchronization 424.
[0172] Synchronization 424 may detect pilot symbols in the 5G symbols to synchronize the transmitted data. In some examples of 5G implementations, pilot symbols may be detected in the time domain at the beginning of a frame (e.g., in a header). Such symbols may be used by the wireless receiver 400 for frame synchronization. In the case of frame synchronization, the 5G symbols proceed to channel estimation 428. Channel estimation 428 may also use time domain pilot symbols and additional frequency domain pilot symbols to estimate time or frequency effects on the received signal (e.g., path loss).
[0173] For example, the channel may be estimated based on the N signals received by the N antennas (except the antenna 404) in the preamble period of each signal. In some examples, the channel estimation 428 may also use the guard interval removed at the guard interval removal 416. Through the channel estimation process, the channel estimation 428 may compensate the frequency domain 5G symbol by a certain factor to minimize the impact on the estimated channel. Although the channel estimation has been described based on the time domain pilot symbol and the frequency domain pilot symbol, other channel estimation techniques or systems (e.g., a MIMO-based channel estimation system or a frequency domain equalization system) are also possible.
[0174] The demodulation map 432 can demodulate the data output from the channel estimate 428. For example, a quadrature amplitude modulation (QAM) demodulator can map the data by changing (e.g., modulating) the amplitude of the associated carrier. Any modulation map described herein can have a corresponding demodulation map as performed by the demodulation map 432. In some examples, the demodulation map 432 can detect the phase of the carrier signal to facilitate demodulation of the 5G symbol. The demodulation map 432 can generate bit data from the 5G symbol for further processing by the deinterleaver 436.
[0175] The deinterleaver 436 may deinterleave the data bits arranged as parity blocks from the demodulation map into the bit stream for the decoder 440. For example, the deinterleaver 436 may perform the inverse operation of convolutional byte interleaving. The deinterleaver 436 may also use channel estimates to compensate for channel effects on the parity blocks.
[0176] Decoder 440 can decode the data output from the scrambler to decode the data. For example, a Reed-Solomon (RS) decoder or a turbo decoder can be used as a decoder to generate a decoded bit stream for descrambler 444. For example, a turbo decoder can implement a parallel serial decoding scheme. In some examples, additional decoding and / or deinterleaving can be performed after decoder 440 and deinterleaver 436. For example, additional decoding can include another decoder that can further decode the data output from decoder 440. Although described in the context of RS decoding and turbo decoding, various decoders 440 are possible, such as a low-density parity check (LDPC) decoder or a polar decoder.
[0177] The descrambler 444 may convert the output data from the decoder 440 from a pseudo-random or random binary sequence to the original source data. For example, the descrambler 44 may convert the decoded data to a transport layer destination (e.g., an MPEG-2 transport stream) which is descrambled with the inverse of the generator polynomial of the scrambler 304. The descrambler thus outputs receiver output data. Thus, the wireless receiver 400 receives an RF transmission including input data X(i,j) 410 to generate receiver output data.
[0178] As in this article, for example, Figure 4 As described, the operation of wireless receiver 400 may include a variety of processing operations. Such operations may be implemented in a conventional wireless receiver, where each operation is implemented by specifically designed hardware for the respective operation. For example, a DSP processing unit may be specifically designed to implement FFT 420. As may be appreciated, additional operations of wireless receiver 400 may be included in a conventional wireless receiver.
[0179] Figure 5 is a schematic diagram of a wireless transmitter 500 according to examples described herein. Figure 5 A wireless transmitter is shown that can perform several operations of an RF front end for wireless transmission with input data X(i,j) 310, including a scrambler 304, a decoder 308, an interleaver 312, a frame adaptation 320, an IFFT 324, a guard interval 326, and an upconversion 328. The transmitter 500 utilizes an electronic device 110 having a processing unit 112 to perform operations of modulation mapping (e.g., modulation mapping 316). For example, the processing unit 112 of the electronic device 110 executes instructions 115 to mix the input data with coefficient data. In the example of the transmitter 500, the input data of the electronic device 110 (e.g., input data 210a-c) can be the output of the interleaver 312; the output data of the electronic device 110 (e.g., output data B(1) 230) can be the input of the frame adaptation 320. For example, the input data of the electronic device 110 may be multiplied with the coefficient data to generate a multiplication result at the multiplication unit / accumulation unit, and the multiplication result may be accumulated at the multiplication unit / accumulation unit to be further multiplied and accumulated with additional coefficients of other portions of the input data and the plurality of coefficients. For example, the processing unit 112 utilizes the selected coefficients so that the coefficients are mixed with the input data to generate a multiplication result reflecting the multiplication result obtained by the multiplication unit / accumulation unit. Figure 2A , 2B Or 2C circuit system uses coefficients to process input data and output data.
[0180] The electronics 110 of the transmitter 500 may retrieve coefficient data specific to the single processing mode selection 502. Figure 5 , the electronic device 110 may receive a single processing mode selection 502. As described herein, a single processing mode may correspond to an aspect of a wireless protocol. Thus, in the example of the transmitter 500, the single mode processing selection 502 may correspond to a modulation mapping aspect of the wireless protocol. Upon receiving this selection 502, the electronic device 110 may execute instructions for implementing the processing mode encoded in the memory 107. For example, the instructions 117 encoded at the memory 107 may include selecting a single processing mode from a plurality of single processing modes, each single processing mode having corresponding coefficient data.
[0181] The single mode processing selection 502 may also indicate an aspect of the wireless protocol for which the electronic device 110 is to execute instructions 115 to generate output data corresponding to the operation of that aspect of the wireless protocol. As depicted, the single processing mode selection 502 indicates that the electronic device 110 is operating as a modulation mapping aspect of the wireless transmitter 500. Thus, the electronic device 110 may implement a modulation mapping processing mode to process input data to retrieve coefficient data corresponding to the selection of the modulation mapping processing mode. When the instructions 115 are executed, the coefficient data may be mixed with the input data of the electronic device 110 to generate output data.
[0182] The single processing mode selection 502 may also indicate the type of modulation mapping. For example, the modulation mapping may be associated with a modulation scheme, including but not limited to: GFDM, FBMC, UFMC, DFDM, SCMA, NOMA, MUSA, or FTN. It may be appreciated that each aspect of the wireless protocol having a corresponding single processing mode may include various types of such aspects, such as modulation mapping processing modes having various modulation schemes selected therefrom.
[0183] The coefficient data corresponding to the selection of the modulation mapping processing mode may be retrieved from a memory (e.g., the memory 145 or a cloud computing database). The coefficients retrieved from the memory are specific to the single processing mode selection 502. In the context of the example of the transmitter 500, the single processing mode selection 502 may indicate that the coefficient data specific to the modulation mapping processing mode is to be retrieved from the memory. Thus, the output data from the electronic device 110 in the transmitter 500 may represent a portion of the transmission of the transmitter processed according to the modulation processing mode selection. The electronic device 110 may output the data such that the frame adaptation 320 receives the output data to further process the transmission.
[0184] Although described above in the context of a modulation mapping processing mode, it should be appreciated that other single processing modes are possible, including but not limited to a fast Fourier transform (FFT) processing mode, an inverse fast Fourier transform (IFFT) processing mode, a decoding mode, a turbo decoding processing mode, a decoding processing mode, a Reed-Solomon processing mode, an interleaver processing mode, a deinterleaving processing mode, a demodulation mapping processing mode, a scrambling processing mode, a descrambling processing mode, a channel estimation processing mode, or a combination thereof. For example, although Figure 5 The electronic device 110 is illustrated as receiving a single processing mode selection 502 to implement the modulation mapping processing mode, but it should be understood that the electronic device 110 may instead Figure 5 The wireless transmitter depicted in FIG. 1 may perform any RF operation on the output data B(u,v) 530 , such that the single processing mode selection 502 indicates that the electronic device 110 may implement another aspect of the wireless transmitter.
[0185] Using this electronic device 110, the transmitter 500 may receive input data X(i,j) 310 to mix with coefficient data specific to aspects of the wireless protocol. For example, the coefficient data may be specific to modulation mapping aspects of the wireless protocol. The transmitter 500 may generate output data B(u,v) 530 that represents the input data processed according to a wireless transmitter (e.g., wireless transmitter 300) configured to operate via the wireless protocol. For example, the output data B(u,v) 530 may correspond to Figure 2A An estimate of B(1)230.
[0186] Figure 6 is a schematic diagram of a wireless transmitter 600 according to examples described herein. Figure 6 A wireless transmitter is shown that can perform several operations performed by an RF front end when wirelessly transmitting input data X(i,j) 310, including a scrambler 304, a frame adaptation 320, an IFFT 324, a guard interval 326, and an up-conversion 328. The transmitter 600 utilizes an electronic device 110 having a processing unit 112 to perform coding, interleaving, and modulation mapping; such as operations of the decoder 308, the interleaver 312, and the modulation mapping 316. For example, the processing unit 112 of the electronic device 110 executes instructions 115 to mix the input data with coefficient data. In the example of the transmitter 600, the input data of the electronic device 110 (e.g., the input data 260a-c) can be the output of the scrambler 304; and the output data of the electronic device 110 (e.g., the output data B(n) 275a-c) can be the input of the frame adaptation 320. For example, the input data of the electronic device 110 may be multiplied with the coefficient data to generate a multiplication result at the multiplication unit / accumulation unit, and the multiplication result may be accumulated at the multiplication unit / accumulation unit to be further multiplied and accumulated with additional coefficients of other portions of the input data and the plurality of coefficients. Thus, the processing unit 112 utilizes the selected coefficients so that the coefficients are mixed with the input data to generate a multiplication result reflecting the multiplication result obtained by the multiplication unit / accumulation unit. Figure 2A to 2B A circuit system uses coefficients to process input data and output data.
[0187] The electronic device 110 of the transmitter 600 may retrieve coefficient data specific to the multi-processing mode selection 602. Figure 6, electronic device 110 receives a multi-processing mode selection 602. As described herein, a multi-processing mode may correspond to at least two aspects of a wireless protocol. Thus, in the example of transmitter 600, multi-processing mode selection 602 corresponds to a coding aspect, an interleaving aspect, and a modulation mapping aspect of the wireless protocol. Upon receiving this selection 602, electronic device 110 may execute instructions for implementing the processing mode encoded in memory 107. For example, instructions 117 encoded at memory 107 may include selecting a multi-processing mode from a plurality of multi-processing modes, each multi-processing mode having respective coefficient data.
[0188] The multi-processing mode selection 602 may also indicate aspects of the wireless protocol for which the electronic device 110 is to execute the instructions 115 to generate output data corresponding to the operation of the aspects of the wireless protocol. As depicted, the multi-processing mode selection 602 indicates that the electronic device 110 is operating as a decoding, interleaving, and modulation mapping of the wireless transmitter 600. Thus, the electronic device 110 may implement a particular multi-processing mode to process the input data to retrieve coefficient data corresponding to the selection of the particular multi-processing mode. When the instructions 115 are executed, the coefficient data may be mixed with the input data of the electronic device 110 to generate output data.
[0189] The multi-processing mode selection 602 may also indicate the type of each aspect of the wireless protocol. For example, as described above, the modulation mapping aspect may be associated with a modulation scheme, including but not limited to: GFDM, FBMC, UFMC, DFDM, SCMA, NOMA, MUSA, or FTN. Continuing in the example, the coding aspect may be associated with a specific type of coding such as RS coding or turbo coding. It should be appreciated that each aspect of the wireless protocol having a corresponding multi-processing mode may include various types of such aspects.
[0190] The coefficient data corresponding to the selection of the multi-processing mode may be retrieved from a memory (e.g., the memory 145 or a cloud computing database). The coefficients retrieved from the memory may be specific to the multi-processing mode selection 602. Thus, the output data from the electronic device 110 in the transmitter 600 may represent a portion of the transmission of the transmitter that is processed according to the multi-processing mode selection 602. The electronic device 110 may output the data such that the frame adaptation 320 receives the output data to further process the transmission.
[0191] Although described above in the context of a particular multi-processing mode selection including coding aspects, interleaving aspects, and modulation mapping aspects, it should be appreciated that other multi-processing modes are possible, including but not limited to any combination of the single processing modes described above. Figure 6The electronic device 110 is illustrated as receiving a multi-processing mode selection 602 to implement a decoding aspect, an interleaving aspect, and a modulation mapping aspect, but it is understood that the electronic device 110 may instead Figure 6 The wireless transmitter depicted in FIG. 1 may perform any RF operation on the output data B(u,v) 630 , such that the single processing mode selection 602 indicates that the electronic device 110 will implement at least two aspects of the wireless transmitter.
[0192] Using this electronic device 110, the transmitter 600 may receive input data X(i,j) 310 to mix with coefficient data specific to at least two aspects of the wireless protocol. The transmitter 600 generates output data B(u,v) 630 that represents the input data processed according to a wireless transmitter (e.g., wireless transmitter 300) configured to operate via the wireless protocol. For example, the output data B(u,v) 630 may correspond to Figure 2A An estimate of the output data B(1) 230.
[0193] Figure 7 is a schematic diagram of a wireless transmitter 700 according to examples described herein. The wireless transmitter 700 may perform several operations of an RF front end for wireless transmission with input data X(i,j) 310. The transmitter 700 utilizes an electronic device 110 having a processing unit 112 to perform the operations of the RF front end. For example, the processing unit 112 of the electronic device 110 may execute instructions 115 to mix the input data with the coefficient data. In the example of the transmitter 700, the output data of the electronic device 110 (e.g., the output data B(n) 275a-c) may be input to the power amplifier 332 after processing the input data X(i,j) 310 (e.g., the input data 210a-c). For example, the input data of the electronic device 110 may be multiplied with the coefficient data to generate a multiplication result at the multiplication unit / accumulation unit, and the multiplication result may be accumulated at the multiplication unit / accumulation unit to be further multiplied and accumulated with other portions of the input data and additional coefficients of the plurality of coefficients. For example, processing unit 112 utilizes selected coefficients so that the coefficients are mixed with input data to generate a signal that reflects the Figure 2A , 2B Or 2C circuit system uses coefficients to process input data and output data.
[0194] The electronics 110 of the transmitter 700 may retrieve coefficient data specific to the full processing mode selection 702. Figure 7, the electronic device 110 may receive a multiple processing mode selection 702. As described herein, a complete processing mode may be a processing mode representing a wireless transmitter (e.g., a wireless transmitter processing mode), or a processing mode representing a wireless receiver (e.g., a wireless receiver processing mode). Thus, in the example of the transmitter 700, the complete processing mode selection 702 may correspond to a wireless transmitter processing mode that may implement any aspects of a specific wireless protocol required for the wireless transmitter to implement the protocol. Similarly, the wireless receiver processing mode may implement any aspects of a specific wireless protocol required for the wireless receiver to implement the protocol. Upon receiving such selection 702, the electronic device 110 may execute instructions 117 for implementing the processing mode encoded in a computer-readable medium. For example, the instructions 117 may include selecting a complete processing mode from a plurality of complete processing modes, each complete processing mode having corresponding coefficient data.
[0195] The full processing mode selection 702 may indicate a wireless protocol for which the electronic device 110 is to execute the instructions 115 to generate output data corresponding to the operation of the wireless protocol as a wireless transmitter. For example, the full processing mode selection 702 may indicate that the wireless transmitter is to implement a 5G wireless protocol including an FBMC modulation scheme. Thus, the electronic device 110 may implement a particular full processing mode to process the input data to retrieve coefficient data corresponding to the selection of the particular full processing mode. When the instructions 115 are executed, the coefficient data may be mixed with the input data of the electronic device 110 to generate output data.
[0196] The coefficient data corresponding to the selection of the full processing mode may be retrieved from a memory (e.g., the memory 145 or a cloud computing database). The coefficients retrieved from the memory are specific to the full processing mode selection 702. Thus, the output data from the electronic device 110 in the transmitter 700 may represent a transmission processed according to the full processing mode selection 702.
[0197] Using this electronic device 110, the transmitter 700 may receive input data X(i,j) 310 to mix with coefficient data of a complete processing mode specific to the wireless protocol. The transmitter 700 may generate output data B(u,v) 730, which represents the input data processed according to a wireless transmitter (e.g., wireless transmitter 300) configured to operate via the wireless protocol. For example, the output data B(u,v) 730 may correspond to Figure 2A An estimate of the output data B(1) 230.
[0198] In some examples, the wireless transmitter mode may include the operation of digital baseband processing, RF front end, and any forward haul processing such as compression or estimation. As an example of forward haul processing, the electronic device 110 may operate in a cloud radio access network (C-RAN), where the wireless base station functions are divided between a remote radio head (RRH) and a baseband unit (BBU). The RRH may perform RF amplification, up / down conversion, filtering, ADC or DAC to provide a baseband signal to the BBU. The BBU may process the baseband signal and optimize resource allocation among the RRHs. The forward haul may be a link between the RRH and the BBU, which may perform compression of the baseband signal to send the signal to the BBU, and may additionally perform estimation of the forward haul link to compensate for any effect of the forward haul on the baseband signal during transmission to the BBU. In such examples, the electronic device 110 may operate as an RRH, a forward haul, a BBU, or any combination thereof by executing instructions 115 to mix input data with coefficient data and / or executing instructions to implement a processing mode such as an RRH processing mode, a forward haul processing mode, or a BBU processing mode.
[0199] In order to train the computing device to generate coefficient data with output data B(u,v)730, the wireless transmitter may receive input associated with RF wireless transmission. Subsequently, the wireless transmitter may perform operations as an RF front end according to the wireless protocol, such as modulating input data for RF wireless transmission. The output data generated by the wireless transmitter may be compared with the input data to generate coefficient data. For example, the comparison may involve a minimization function that optimizes the coefficients. The minimization function analyzes each combination of input samples and test vectors to determine a minimum amount, which may reflect the minimized error of the difference between the two outputs. The coefficients used to generate the minimized amount reflect the minimum error estimate of the output from the specially designed wireless transmitter. Therefore, the electronic device 110 can be trained to generate coefficient data based on the operation of the wireless transmitter, so that mixing arbitrary input data (e.g., test vectors) using the coefficient data generates an approximation of the output data, as if it is processed by the specially designed wireless transmitter. The coefficient data may also be stored in a coefficient database (e.g., memory 145), where each coefficient data set corresponds to a specific wireless protocol in the RF domain that can be used for data transmission. In some examples, input test vectors and output test vectors that simulate the processing of the wireless transmitter can be used to generate coefficient data.
[0200] Although described herein in the context of a wireless transmitter having a single processing mode selection (e.g., Figure 5 ), multiple processing mode selection (e.g., as depicted in Figure 6 ), and complete processing mode selection (e.g., as Figure 7 ), but in some instances, such as by instructions 117 for implementing a wireless processing mode, the processing mode may be performed at a wireless receiver (e.g., Figure 4 10) is implemented in a wireless receiver 400 of the present invention, wherein the corresponding aspects of the wireless protocol are implemented by an electronic device 110 having a processing unit 112. For example, a computing system may receive wireless input data from an antenna 404 by an electronic device 110 implementing a wireless receiver processing mode, as indicated by a complete processing mode selection indicating a wireless receiver processing mode. Thus, whether a single aspect, multiple aspects, or a complete implementation, any aspect of a wireless transmitter or wireless receiver may be implemented by an electronic device 110 having a processing unit 112 that implements instructions 115 for mixing input data with coefficient data and instructions 117 for implementing a processing mode. Similarly, an electronic device 110 may be trained by coefficient data generated from a wireless receiver configured to operate according to a wireless protocol. For example, output data generated by a wireless receiver (or any aspect of a wireless receiver) may be compared to input data to generate coefficient data.
[0201] Additionally or alternatively, any processing mode selection may indicate the type of each aspect of the wireless protocol. For example, a modulation mapping aspect may be associated with a modulation scheme, including but not limited to: GFDM, FBMC, UFMC, DFDM, SCMA, NOMA, MUSA, or FTN. Each aspect of the wireless protocol having a corresponding processing mode may include various types of such aspects. For example, full processing mode selection 702 may indicate aspects for up-conversion to a 5G wireless frequency range, such as utilizing a carrier frequency in an E-band (e.g., 71 to 76 GHz and 81 to 86 GHz), a 28 GHz millimeter wave (mmWave) band, or a 60 GHz V-band (e.g., implementing an 802.11 AD protocol). In some examples, full processing mode selection 702 may also indicate whether a MIMO implementation will be utilized, such as selection 702 may indicate that 2×2 spatial streams may be utilized.
[0202] Figure 8 is a flow chart of a method 800 according to examples described herein. Figure 1A System 100 in Figure 2A System 200, Figure 2B System 250, Figure 2C The system 280 described herein Figures 3 to 7 or Figures 10 to 11 The example method 800 can be implemented in any system or combination of the systems depicted in the example method 800. In some examples, the blocks in the example method 800 can be implemented by, for example, Figure 1A The computing device of the electronic device 110 and / or in combination with, for example Figure 2A , 2BThe operations described in blocks 804-836 may also be stored as computer-executable instructions in a computer-readable medium, such as a mode configurable control 105 storing executable instructions 115.
[0203] The example method 800 may begin at block 804, which begins performing a mix of input data with coefficient data routine. The method may include block 808, which states "obtain a processing mode selection corresponding to a processing mode of a processing unit". As described herein, the processing mode selection may be obtained as a selection from a touch screen of a computing device in communication with the electronic device 110. The processing mode selection may be obtained from any computing device that can receive user input regarding the processing mode selection. Block 808 may be followed by block 812, which states "configure the processing unit to select a plurality of coefficients based on the processing mode selection". As described herein, configuring the processing unit may include configuring the processing unit for various processing modes depending on the processing mode selection. Configuring the processing unit may include configuring the processing unit for various modes including a single processing mode, a multi-processing mode, and a full processing mode. For example, the computing system may operate as a turbo decoding operation in a single processing mode to output data according to data being encoded by the turbo decoding operation.
[0204] Block 812 may be followed by block 816, which recites "Retrieve a plurality of coefficients from memory." As described herein, the processing unit may retrieve coefficients for mixing with the input data; for example, utilizing a memory lookup unit. For example, the memory may store (e.g., in a database) correlations between coefficients and wireless protocols and / or processing modes described herein. For example, the processing unit may request the coefficients from a memory portion of an implementing computing device, from a memory portion of an external computing device, or from a memory implemented in a cloud computing device. In turn, the plurality of coefficients may be retrieved from memory as requested by the processing unit.
[0205] Block 816 may be followed by block 820, which recites "Obtaining input data for transmission in a radio frequency (RF) domain based on a processing mode selection." As described herein, a processing unit may be configured to obtain a plurality of types of input data, such as data bit streams, coded bits, modulation symbols, etc., that may be transmitted or received by a wireless transmitter or a wireless receiver, respectively. In an example, a processing unit in a single processing mode may implement the functionality of an operation of a portion of a wireless transmitter or a receiver. In an example of a turbo decoding operation, a processing unit may obtain a data bit stream to be decoded, including an indication of parameters regarding a turbo decoding operation. In an example of an IFFT operation, a processing unit may obtain a data bit stream to be transformed into a frequency domain, including an indication of parameters regarding an IFFT operation, such as a point size for an IFFT operation to be utilized. In an example of a multi-processing mode operation of a DAC, the input data may be digital data that is converted into analog data for transmission at an antenna of a wireless transmitter. In some examples, parameters regarding aspects of the operation may be obtained in a processing mode selection. For example, parameters of an IFFT operation, such as a point size used for the IFFT operation, may be obtained as information included in a single processing mode selection, a multi-processing mode selection, or a full processing mode selection.
[0206] Block 820 may be followed by block 824, which recites “Compute, at a first multiplication / accumulation processing unit (MAC unit) layer, a delayed version of the input data and a corresponding output of the first MAC unit layer using a plurality of coefficients to generate a first processing result.” As described herein, the processing unit utilizes a plurality of coefficients such that the coefficients are mixed with the input data and the delayed version of the corresponding output of the first MAC unit layer to generate a first processing result reflecting the processing result generated by the first MAC unit. Figure 2A , 2B A circuit system of a 2C circuit may process input data using coefficients to output data. For example, various ALUs in an integrated circuit may be configured as Figure 2A , 2B or 2C circuit system operation, thereby mixing the input data and the delayed version of the corresponding output of the first MAC unit layer with the coefficients as described herein. For example, referring to Figure 2A , at a first multiplication / accumulation processing unit (MAC unit) layer, a delayed version of the input data and the corresponding output of the first MAC unit layer is calculated using a plurality of coefficients to generate a first processing result. In some examples, various hardware platforms may implement Figure 2A , 2B Or a circuit system of 2C, such as an ASIC, a DSP implemented as part of an FPGA, or a system on a chip.
[0207] Block 824 may be followed by block 828, which recites “computing the first processing result and the delayed version of at least a portion of the first processing result with an additional plurality of coefficients at the additional MAC unit layer to generate a second processing result.” As described herein, the processing unit utilizes the additional plurality of coefficients such that blending the coefficients with the particular processing results and the delayed version of at least a portion of those particular processing results generates a second processing result that reflects the processing result generated by the processing unit. Figure 2A , 2B or 2C circuit system uses coefficients to process input data and output data. For example, refer to Figure 2A At the second multiplication / accumulation processing unit (MAC unit) layer, the processing results of the first layer (e.g., multiplication processing results) and delayed versions of at least a portion of those processing results may be calculated using an additional plurality of coefficients to generate a second processing result. The processing results of the second layer may be calculated using an additional plurality of coefficients to generate output data B(1) 230.
[0208] Block 828 may be followed by block 832, which states, "Provide output data representing the portion of the transmission processed in accordance with the processing mode selection." As described herein, the output data may be provided to another portion of the specially designed hardware, such as another portion of the wireless transmitter or receiver, or an antenna for wireless RF transmission. In the example of a full processing mode selection, the output data may be provided to an application that requested the output data from the wireless endpoint, such that the output data is provided to the application as part of the processing unit implementing the wireless receiver. Block 832 may be followed by block 836, which ends the example method 800. In some examples, blocks 808 and 812 may be optional blocks.
[0209] Fig. 9 is a flow chart of a method 900 according to examples described herein. Example method 900 may use, for example, Figure 1A System 100 in Figure 2A System 200, Figure 2B System 250, Figure 2C The system 280 described herein Figures 3 to 7 or Figures 10 to 11 In some examples, the blocks in the example method 900 may be implemented by, for example, Figure 1A The computing device of the electronic device 110 and / or the combination Figure 2A , 2B The operations described in blocks 904 to 928 may also be stored as computer executable instructions in a computer readable medium, such as a mode configurable control 105 storing executable instructions 115, or a memory 107 storing executable instructions 117.
[0210] The example method 900 may begin at block 904, which begins to perform a mix input data with coefficient data routine. The method may begin at block 908, which states "Obtain input data associated with a radio frequency (RF) wireless transmission." As described herein, a processing unit may be configured to obtain a plurality of types of input data, such as data bit streams, coded bits, modulation symbols, etc., which may be transmitted or received by a wireless transmitter or a wireless receiver, respectively. In an example, a processing unit in a single processing mode may implement the functions of the operation of a portion of a wireless transmitter or a receiver. In an example of a turbo decoding operation, the processing unit may obtain a data bit stream to be decoded, including an indication of parameters for the turbo decoding operation. In an example of an IFFT operation, the processing unit may obtain a data bit stream to be transformed into a frequency domain, including an indication of parameters for the IFFT operation, such as a point size for the IFFT operation to be utilized. In an example of a multi-processing mode operation of a DAC, the input data may be digital data that is converted into analog data for transmission at an antenna of a wireless transmitter. In some examples, parameters regarding aspects of the operation may be received in a processing mode selection. For example, parameters of an IFFT operation, such as a point size used for the IFFT operation, may be obtained as information included in a single processing mode selection, a multi-processing mode selection, or a full processing mode selection.
[0211] Block 908 may be followed by block 912, which recites "Performing operations of the RF front end on the input data to modulate the input data for RF wireless transmission." As described herein, a specially designed wireless transmitter may perform the RF front end operations described above, such as scrambling, coding, interleaving, modulation mapping, up-conversion, etc. to transmit the input data as an RF wireless transmission. In an example, the specially designed wireless transmitter may perform only a portion of the RF front end operations to train a computing device for a particular operation of the wireless transmitter so that coefficient data generated by the computing device may be used in a single processing mode or a multi-processing mode.
[0212] Block 912 may be followed by block 916, which recites "Comparing the output data to the input data to generate coefficient data." As described herein, the computing device may compare the output data from the specially designed wireless transmitter to the output data from the processing unit of the computing device implementing the operation of the wireless transmitter. For example, the computing device may use a minimum error p-norm comparison between the two outputs as part of a minimization function to determine whether the coefficient data represents an estimate of the specially designed wireless transmitter. In an example, the computing device may use a minimum square error function between the two outputs as part of an optimization problem to determine whether the coefficient data represents an estimate of the specially designed wireless transmitter.
[0213] As an example of this comparison, Figure 2A , 2BThe processing unit 112 of 2C may be referred to as a neuron, where a number of similar processing units 112 operate as a neural network. Figure 2A In the case where such a processing unit 112 operates as a neuron, the processing unit 112 may receive an observation vector Y(n) as input data 210a-c. For example, the observation vector may be represented as:
[0214] Y(n)=[y(n),y(n-1),...y(n-m+1)]
[0215] And the coefficients stored in the memory 145 may be referred to as connection weights, which may be randomly initialized to specific values so that for Figure 2A The MLU of the retrieval coefficient can be expressed as:
[0216] and (for i=1,2,...m; j=1,2,...M)
[0217] Where m represents the number of antennas or wireless channels, and where M represents the number of layers. When such observation vectors are provided as inputs and connection weights as coefficients, an electronic device implementing such a processing unit may compare output data generated from the processing unit to obtain trained coefficient data, or connection weights.
[0218] Block 916 may be followed by block 920, which states, "Training the computing device to generate coefficient data based on operations performed on the input data such that the input data and a delayed version of at least a portion of the first processing result are mixed using the coefficient data to generate an approximation of the output data." As described herein, the computing device may compare output data from a specially designed wireless transmitter with output data from a processing unit of the computing device implementing operations of the wireless transmitter across various test vectors and input samples to train the computing device to determine minimized error. For example, the minimized error trained across various input samples and test vectors may represent an optimized estimate of the processing of the input data in the specially designed hardware. In an example, training the computing device according to the input samples and test vectors may be referred to as supervised learning of the computing device to generate coefficient data. In various examples, the computing device may also be trained to generate coefficient data according to unsupervised learning. For example, the computing device may monitor the output of the specially designed hardware to learn coefficient data that may minimize the error of the processing unit of the computing device implementing operations of the wireless transmitter.
[0219] Continuing with the example of the observation vector and connection weights, processing unit 112 may repeat the comparison of the generated output data with the input data such that the connection weights are updated in each iteration to minimize the error of the output of the neuron (e.g., output B(1) 230). When multiplied and accumulated with this observation vector as Y(n), the connection weights may be updated according to equation (19):
[0220]
[0221] In equation (19), α is the learning rate at which processing unit 112 can learn from the output data and thereby calculate updated connection weights. may refer to the error term that is propagated to the next set of connection weights to be updated. For example, the error term may be the minimized error relative to the p-norm calculated by comparing the output data B(1) 230 with the particular input data 210a-c. Alternatively, relative to Figure 2B In the example of , the error term may be the minimized error relative to the p-norm calculated by comparing the output data B(n) 275a-c with the particular input data 260a-c. Thus, using equation (19), coefficient data representing the minimized error between the output data and the input data may be generated. Thus, when a computing device is trained to generate coefficient data based on operations performed on input data, such that the approximation of the output data is generated by mixing the input data using the coefficient data, a system disclosed herein, such as Figures 3 to 7 Or any number of multiple coefficient data utilized by a 10 to 11 system.
[0222] Block 920 may be followed by block 924, which recites "Store coefficient data in a coefficient memory." As described herein, the computing device may store the coefficients in a memory, such as a memory database. For example, the memory may store correlations between the coefficients and the wireless protocol and / or processing mode as trained by the computing device. For example, the computing device may store the coefficients in a memory portion of the computing device itself, in a memory of an external computing device, or in a memory implemented in a cloud computing device. Block 924 may be followed by block 928, which ends the example method 900. In some examples, block 924 may be an optional block.
[0223] The blocks included in the described example methods 800 and 900 are for illustrative purposes. In some embodiments, the blocks may be executed in different orders. In some other embodiments, various blocks may be eliminated. In yet other embodiments, various blocks may be divided into additional blocks, supplemented with other blocks, or combined together into fewer blocks. Other variations of these specific blocks are contemplated, including changes in the order of the blocks, changes in the contents of the blocks that are divided or combined into other blocks, etc.
[0224] Fig.10An example of a wireless communication system 1000 according to aspects of the present disclosure is illustrated. The wireless communication system 1000 includes a base station 1010, a mobile device 1015, a drone 1017, a small cell 1030, and vehicles 1040, 1045. The base station 1010 and the small cell 1030 can be connected to a network that provides access to the Internet and traditional communication links. The system 1000 can facilitate a wide range of wireless communication connections in a 5G system that can include various frequency bands, including but not limited to: sub-6 GHz frequency bands (e.g., 700 MHz communication frequency), mid-range communication frequency bands (e.g., 2.4 GHz), millimeter wave frequency bands (e.g., 24 GHz), and NR frequency bands (e.g., 3.5 GHz).
[0225] Additionally or alternatively, the wireless communication connection may support various modulation schemes, including but not limited to: filter bank multi-carrier (FBMC), generalized frequency division multiplexing (GFDM), universal filtered multi-carrier (UFMC) transmission, dual orthogonal frequency division multiplexing (BFDM), sparse code multiple access (SCMA), non-orthogonal multiple access (NOMA), multi-user shared access (MUSA), and super-Nyquist (FTN) signaling with time-frequency packing. Such frequency bands and modulation techniques may be part of a standard framework such as Long Term Evolution (LTE) (e.g., 1.8 GHz band) or other technical specifications published by organizations such as 3GPP or IEEE, which may include various specifications for subcarrier frequency ranges, number of subcarriers, uplink / downlink transmission speeds, TDD / FDD, and / or other aspects of wireless communication protocols.
[0226] System 1000 may depict aspects of a radio access network (RAN), and system 1000 may communicate with or include a core network (not shown). The core network may include one or more serving gateways, mobility management entities, home subscriber servers, and packet data gateways. The core network may facilitate user and control plane links to mobile devices via the RAN, and it may be an interface to external networks (e.g., the Internet). Base stations 1010, communication devices 1020, and small cells 930 may be coupled to the core network or to each other, or both, via wired or wireless backhaul links (e.g., S1 interfaces, X2 interfaces, etc.).
[0227] The system 1000 can provide a communication link connected to a device or "thing", such as a sensor device (e.g., solar cell 1037), to provide an Internet of Things ("IoT") framework. Things connected within the IoT can operate within a frequency band licensed to and controlled by a cellular network service provider, or such devices or things can do so. Such frequency bands and operations can be referred to as narrowband IoT (NB-IoT) because the frequency bands allocated for IoT operations can be small or narrow relative to the overall system bandwidth. For example, the frequency bands allocated for NB-IoT can have a bandwidth of 1, 5, 10, or 20 MHz.
[0228] Additionally or alternatively, the IoT may include devices or things that operate at frequencies different from traditional cellular technologies to facilitate the use of wireless spectrum. For example, the IoT framework may allow multiple devices in the system 1000 to operate at a frequency band below 6 GHz or other industrial, scientific and medical (ISM) radio bands where devices can operate on unlicensed shared spectrum. The frequency band below 6 GHz may also be characterized as and may also be characterized as the NB-IoT frequency band. For example, when operating at a low frequency range, a device that provides sensor data for a "thing" such as a solar cell 1037 may utilize less energy, thereby obtaining power efficiency and may utilize a less complex signaling framework so that the device can transmit asynchronously at a frequency band below 6 GHz. The frequency band below 6 GHz may support a wide variety of use cases, including the transmission of sensor data from various sensor devices. Examples of sensor devices include sensors for detecting energy, heat, light, vibration, biosignals (e.g., pulse, EEG, EKG, heart rate, respiratory rate, blood pressure), distance, speed, acceleration, or a combination thereof. Sensor devices may be deployed in buildings, on individuals, and / or other locations in the environment. The sensor devices may communicate with each other and with a computing system, which may aggregate and / or analyze data provided from one or more sensor devices in the environment.
[0229] In this 5G framework, the device may perform functionality performed by base stations in other mobile networks (e.g., UMTS or LTE), such as forming connections between nodes or managing mobility operations (e.g., handoff or reselection). For example, the mobile device 1015 may utilize the mobile device 1015 to receive sensor data, such as blood pressure data, from a user and may transmit the sensor data to the base station 1010 over the narrowband IoT band. In this example, some parameters for determination by the mobile device 1015 may include the availability of licensed spectrum, the availability of unlicensed spectrum, and / or the time-sensitive nature of the sensor data. Continuing in the example, the mobile device 1015 may transmit the blood pressure data because the narrowband IoT band is available and may transmit the sensor data quickly, thereby identifying a time-sensitive component to the blood pressure (e.g., where the blood pressure measurement is dangerously high or low, such as where the systolic blood pressure is three standard deviations away from the norm).
[0230] Additionally or alternatively, the mobile device 1015 may form a device-to-device (D2D) connection with other mobile devices or other elements of the system 1000. For example, the mobile device 1015 may form an RFID, WiFi, MultiFire, Bluetooth, or ZigBee connection with other devices including the communication device 1020 or the vehicle 1045. In some examples, a licensed spectrum band may be used for D2D connection, and such connection may be managed by a cellular network or service provider. Therefore, although the above examples are described in the context of narrowband IoT, it can be appreciated that the mobile device 1015 may utilize other device-to-device connections to provide information (e.g., sensor data) collected on a frequency band different from the frequency band determined by the mobile device 1015 for transmission of the information.
[0231] In addition, some communication devices can facilitate self-organizing networks (e.g., networks formed with communication devices 1020) attached to stationary objects and vehicles 1040, 1045, without necessarily forming traditional connections to base stations 1010 and / or core networks. Other stationary objects can be used to support communication devices 1020, such as, but not limited to, trees, plants, poles, buildings, airships, dirigibles, balloons, street signs, mailboxes, or combinations thereof. In this system 1000, communication devices 1020 and small cells 1030 (e.g., small cells, femtocells, WLAN access points, cellular hotspots, etc.) can be mounted on or attached to another structure, such as lamp posts and buildings, to facilitate the formation of self-organizing networks and other IoT-based networks. Such networks can operate at different frequency bands than the prior art, such as mobile devices 1015 communicating with base stations 1010 on cellular communication bands.
[0232] The communication device 1020 may form a wireless network that operates in a hierarchical or ad hoc network manner, depending in part on a connection to another element of the system 1000. For example, the communication device 1020 may utilize a 700 MHz communication frequency in an unlicensed spectrum to form a connection with the mobile device 1015 while utilizing a licensed spectrum communication frequency to form another connection with the vehicle 1045. The communication device 1020 may communicate with the vehicle 1045 on the licensed spectrum to provide direct access to time-sensitive data, such as data for autonomous driving capabilities of the vehicle 1045 on the 5.9 GHz band of dedicated short-range communications (DSRC).
[0233] The vehicles 1040 and 1045 may form an ad hoc network at a different frequency band than the connection between the communication device 1020 and the vehicle 1045. For example, for a high bandwidth connection for providing time-sensitive data between the vehicles 1040, 1045, the 24 GHz millimeter wave band may be used for data transmission between the vehicles 1040, 1045. For example, while the vehicles 1040, 1045 are crossing a narrow intersection with each other, the vehicles 1040, 1045 may share real-time direction and navigation data with each other over the connection. Each vehicle 1040, 1045 may track the intersection and provide image data to an image processing algorithm to facilitate autonomous navigation of each vehicle while each is traveling along the intersection. In some examples, this real-time data may also be substantially simultaneously shared over an exclusive licensed spectrum connection between communication device 1020 and vehicle 1045, for example, to process image data received at both vehicle 1045 and vehicle 1040, such as image data transmitted from vehicle 1040 to vehicle 1045 over the 24 GHz millimeter wave band. Fig.10 An automobile is shown in FIG. 1 , but other vehicles may be used, including but not limited to an aircraft, a spacecraft, a hot air balloon, an airship, a dirigible, a train, a submarine, a ship, a ferry, a cruise ship, a helicopter, a motorcycle, a bicycle, a drone, or a combination thereof.
[0234] While communication device 1020 may utilize stationary objects to support and estimate input data received at such device, signals transmitted to / from vehicles 1040 and 1045 may experience time-varying wireless channels due to time-varying objects in the signal path to / from vehicles 1040 and 1045, such as objects on the road, pedestrians or other vehicles, and / or moving objects. Advantageously, utilizing Figures 2A to 2C1040 and 1045. For example, when noisy wireless input data is received and this input data is mixed with a delayed version of the processing results, the vehicles 1040 and 1045 can more effectively process and compensate for the effects of time-varying objects in the path of the vehicles 1040 and 1045. For example, while one frequency band of the vehicle 1040 may operate in the ultra-wideband (UWB) range to detect nearby objects, and this effect may be compensated independently of another frequency band received on the 24 GHz millimeter wave band or even the cellular band. Time-varying objects may affect the frequencies of each frequency band with different effects, so each frequency band may be processed independently, but compensation may also be made based on the time-varying effects of the corresponding frequencies using the recurrent neural network described herein.
[0235] Although described in the context of the 24 GHz millimeter wave band, it can be appreciated that connections can be formed in the system 1000 in other millimeter wave bands or other frequency bands that can be licensed or unlicensed bands, such as 28 GHz, 37 GHz, 38 GHz, 39 GHz. In some cases, the vehicles 1040, 1045 can share the frequency bands in which they communicate with other vehicles in different networks. For example, a convoy can pass by the vehicle 1040 and temporarily share the 24 GHz millimeter wave band to form a connection between the convoys in addition to the 24 GHz millimeter wave connection between the vehicles 1040, 1045. As another example, the communication device 1020 can substantially simultaneously maintain a 700 MHz connection with a mobile device 1015 operated by a user (e.g., a pedestrian walking along a street) to provide information about the user's location to the vehicle 1045 on the 5.9 GHz band. In providing such information, the communication device 1020 may take advantage of antenna diversity schemes as part of a massive MIMO framework to facilitate time-sensitive separate connections with both the mobile device 1015 and the vehicle 1045. A massive MIMO framework may involve transmitting and / or receiving devices having a large number of antennas (e.g., 12, 20, 64, 128, etc.), which may facilitate precise beamforming or spatial diversity that is unattainable with devices operating with fewer antennas according to traditional protocols (e.g., WiFi or LTE).
[0236] The base station 1010 and the small cell 1030 may wirelessly communicate with devices in the system 1000 or other communication-capable devices in the system 1000 that have at least a sensor wireless network (e.g., a solar cell 1037 that may operate on an active / sleep cycle, and / or one or more other sensor devices). The base station 1010 may provide wireless communication coverage for devices that enter its coverage area, such as mobile devices 1015 and drones 1017. The small cell 1030 may provide wireless communication coverage for devices that enter its coverage area, such as vehicles 1045 and drones 1017, for example, near a building where the small cell 1030 is installed.
[0237] In general, small cells 1030 may be referred to as small cells and provide coverage for a local geographic area, e.g., 200 meters or less in some instances. This may be contrasted with macro cells, which may provide coverage over a wide or large area of approximately several square miles or kilometers. In some instances, small cells 1030 may be deployed (e.g., mounted on a building) within some coverage areas of base stations 1010 (e.g., macro cells) where wireless communication traffic may be dense based on traffic analysis of the coverage areas. For example, if base station 1010 typically receives and / or transmits a higher volume of wireless communication transmissions than other coverage areas of the base station 1010, the small cell 1030 may be deployed (e.g., mounted on a building) within some coverage areas of the base station 1010 (e.g., macro cells) where wireless communication traffic may be dense ... Fig.10 The base stations 1010 are deployed on buildings in the coverage area of the base station 1010. The base stations 1010 may be deployed in a geographic area, thereby providing wireless coverage for a portion of the geographic area. As wireless communication traffic becomes more intensive, additional base stations 1010 may be deployed in certain areas, which may change the coverage area of existing base stations 1010, or other supporting stations, such as small cells 1030, may be deployed. The small cell 1030 may be a femtocell, which may provide coverage for an area smaller than a small cell, for example, 100 meters or less (e.g., one floor of a building) in some instances.
[0238] Although base station 1010 and small cell 1030 may provide communication coverage for a portion of a geographic area surrounding their respective areas, both may vary aspects of their coverage to facilitate faster wireless connections for particular devices. For example, small cell 1030 may provide coverage primarily for devices around or in a building where small cell 1030 is installed. However, small cell 1030 may also detect devices that have entered the coverage area and adjust its coverage area to facilitate faster connections to such devices.
[0239] For example, the small cell 1030 may support a large MIMO connection with a drone 1017, which may also be referred to as an unmanned aerial vehicle (UAV), and when a vehicle 1045 enters its coverage area, the small cell 1030 adjusts some antennas to point directionally in the direction of the vehicle 1045 rather than the drone 1017 to facilitate a large MIMO connection with the vehicle in addition to the drone 1017. When some antennas are adjusted, the small cell 1030 may not be able to support a connection to the drone 1017 at a particular frequency as fast as before the adjustment. For example, the small cell 1030 may communicate with the drone 1017 at a first frequency among various possible frequencies in the 4G LTE band of 1.8 GHz. However, the drone 1017 may also request to connect with another device (e.g., the base station 1010) in its coverage area at a different frequency, which may facilitate a similar connection as described with reference to the small cell 1030, or a different (e.g., faster, more reliable) connection with the base station 1010, for example, at a frequency of 3.5 GHz in the 5G NR band. Thus, the system 1000 can enhance an existing communication link while providing additional connectivity to devices that may utilize or require such a link, such that input data is mixed with coefficient data to generate output data using a processing unit; for example, output data to be transmitted over 4GE LTE and 5G NR bands. In some examples, the drone 1017 can act as a mobile or aerial base station.
[0240] Similar to signals transmitted to / from vehicles 1040 and 1045 that may experience time-varying wireless channels due to time-varying objects in the signal paths to / from vehicles 1040 and 1045, multiple drones 1017 may also experience this time-varying effect (e.g., in a drone network). Figures 2A to 2C , the time-varying properties of moving aerial objects can be better estimated by an inference recurrent neural network trained to process RF signals in such time-varying wireless channels. For example, when receiving noisy wireless input data and mixing this input data with delayed versions of the processing results, multiple drones 1017 can more effectively process and compensate for the effects of time-varying objects in their respective aerial paths, including on different frequency bands, such as in the case of multiple wireless connections for a single drone 1017 as described above.
[0241] The wireless communication system 1000 may include devices such as a base station 1010, a communication device 1020, and a small cell 1030, which may support multiple connections at different frequencies to devices in the system 1000, so that input data is mixed with coefficient data to utilize processing units implemented in electronic devices of the system 1000, e.g. Figure 1A , 2A, 2B, or 2C generates output data. Such devices may operate in a hierarchical mode or an ad hoc mode with other devices in the network of system 1000. Although described in the context of base station 1010, communication device 1020, and small cell 1030, it can be appreciated that other devices that can support several connections with devices in the network such that input data is mixed with coefficient data to generate output data using processing unit 112 may be included in system 1000, including but not limited to: macro cells, femto cells, routers, satellites, and RFID detectors.
[0242] In various examples, elements of the wireless communication system 1000, such as the base station 1010, the mobile device 1015, the drone 1017, the communication device 1020, the small cell 1030, and the vehicles 1040, 1045, can be implemented as electronic devices described herein that mix input data with coefficient data to generate output data using the processing unit 112. For example, the communication device 1020 can be implemented as an electronic device described herein, such as Figure 1A The electronic device 110 implements Figure 2A , 2B Or processing unit 112 of 2C, or any system or combination of systems depicted in the figures described herein.
[0243] Fig.11 An example of a wireless communication device 1100 according to aspects of the present disclosure is illustrated. The wireless communication device 1100 includes a mobile device 1115, a drone 1117, a communication device 1120, and a small cell 1130. The building 1110 also includes devices of the wireless communication device 1100, which can be configured to communicate with other elements in the building 1110 or the small cell 1130. The building 1110 includes networked workstations 1140, 1145, a virtual reality device 1150, IoT devices 1155, 1160, and a networked entertainment device 1065. In the depicted device 1100, the IoT devices 1155, 1160 can be a washing machine and a dryer for residential use, respectively, controlled by the virtual reality device 1150. Thus, while a user of the virtual reality device 1150 can be in a different room of the building 1110, the user can control the operation of the IoT device 1155, such as configuring washing machine settings. The virtual reality device 1150 can also control the networked entertainment device 1065. For example, the virtual reality device 1150 may broadcast a virtual game being played by a user of the virtual reality device 1150 to a display of the networked entertainment device 1065 .
[0244] Similar to the signals transmitted to / from vehicles 1040 and 1045 that may experience a time-varying wireless channel due to time-varying objects in the signal path to / from vehicles 1040 and 1045, the user of virtual reality device 1150 may also experience this time-varying effect in building 1110. Advantageously, utilizing Figures 2A to 2C 110 , the time-varying properties of moving persons or objects may be better estimated by an inference recurrent neural network trained to process RF signals in such time-varying wireless channels. For example, virtual reality device 1150 may more effectively process and compensate for the effects of time-varying objects (e.g., networked IoT devices such as robotic vacuum devices) or persons (e.g., other persons in the room above or below the user of virtual reality device 1150) in building 1110 when receiving noisy wireless input data and mixing this input data with a delayed version of the processing results.
[0245] Any of the devices of the small cell 1130 or the building 1110 may be connected to a network that provides access to the Internet and traditional communication links. Similar to the system 1000, the device 1100 may facilitate a wide range of wireless communication connections in a 5G system that may include various frequency bands, including but not limited to: sub-6 GHz frequency bands (e.g., 700 MHz communication frequency), mid-range communication frequency bands (e.g., 2.4 GHz), and millimeter wave frequency bands (e.g., 24 GHz). Additionally or alternatively, the wireless communication connection may support various modulation schemes as described above with reference to the system 1000. The device 1100 is operable and configured to communicate similarly to the system 1000. Thus, similarly numbered elements of the device 1100 and the system 1000 may be configured in a similar manner, such as the communication device 1020 and the communication device 1120, the small cell 1030 and the small cell 1130, etc.
[0246] Similar to system 1000, in which the elements of system 1000 are configured to form independent hierarchical or self-organizing networks, the communication device 1120 can form a hierarchical network with the small cell 1130 and the mobile device 1115, while an additional self-organizing network can be formed within the small cell 1130 network including the drone 1117 and some devices of the building 1110 (e.g., networked workstations 1140, 1145 and IoT devices 1155, 1160).
[0247] Devices in the communication device 1100 may also form (D2D) connections with other mobile devices or other elements of the system 1100. For example, the virtual reality device 1150 may form narrowband IoT connections with other devices including the IoT device 1155 and the networked entertainment device 1065. As described above, in some examples, licensed spectrum bands may be used for D2D connections, and such connections may be managed by the cellular network or service provider. Thus, while the above examples are described in the context of narrowband IoT, it may be appreciated that the virtual reality device 1150 may utilize other device-to-device connections.
[0248] In various examples, elements of the wireless communication device 1100, such as the mobile device 1115, the drone 1117, the communication device 1120 and the small cell 1130, the networked workstations 1140, 1145, the virtual reality device 1150, the IoT devices 1155, 1160, and the networked entertainment device 1065 may be implemented as electronic devices described herein that mix input data with coefficient data to generate output data using the processing unit 112. For example, the communication device 1120 may be implemented as an electronic device described herein, such as Figure 1A The electronic device 110 implements Figure 2A , 2B Or processing unit 112 of 2C, or any system or combination of systems depicted in the figures described herein.
[0249] Certain details are set forth above to provide a full understanding of the described examples. However, it will be understood by those skilled in the art that the examples may be practiced without various of these specific details. The description set forth herein in conjunction with the accompanying drawings describes example configurations and does not represent all examples that may be implemented or within the scope of the claims. The terms "exemplary" and "example" as used herein refer to "serving as an example, instance, or illustration" and are not "preferred" or "more advantageous than other examples". The detailed description includes specific details for the purpose of providing an understanding of the described techniques. However, these techniques may be practiced without these specific details. In some examples, well-known structures and devices are shown in the form of block diagrams in order to avoid confusing the concepts of the described examples.
[0250] The information and signals described herein may be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips that may be referenced throughout the above description may be represented by voltage, current, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof.
[0251] The techniques described herein may be used in various wireless communication systems, which may include multiple access cellular communication systems, and which may employ code division multiple access (CDMA), time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal frequency division multiple access (OFDMA), or single carrier frequency division multiple access (SC-FDMA), or any combination of such techniques. Some of these techniques have been used in or are related to standardized wireless communication protocols of organizations such as the 3rd Generation Partnership Project (3GPP), the 3rd Generation Partnership Project 2 (3GPP2), and the IEEE. These wireless standards include Ultra Mobile Broadband (UMB), Universal Mobile Telecommunications System (UMTS), Long Term Evolution (LTE), LTE-Advanced (LTE-A), LTE-A Pro, New Radio (NR), IEEE 802.11 (WiFi), and IEEE 802.16 (WiMAX), among others.
[0252] The term "5G" or "5G communication system" may refer to a system operating according to standardized protocols developed or discussed after, for example, LTE Release 13 or 14 or WiMAX 802.16e-2005 by their respective sponsoring organizations. The features described herein may be employed in systems configured according to other generations of wireless communication systems, including those configured according to the standards described above.
[0253] The various illustrative blocks and modules described in conjunction with the disclosure herein may be implemented or executed using a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array signal (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, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices (e.g., 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).
[0254] The functions described herein may be implemented in hardware, software executed by a processor, firmware, or any combination thereof. If implemented in software executed by a processor, the functions may be stored on a computer-readable medium or transmitted via a computer-readable medium as one or more instructions or codes. Computer-readable media include both non-transitory computer storage media and communication media, and the communication media include any media that facilitates the transfer of a computer program from one place to another. Non-transitory storage media may be any available media that can be accessed by a general-purpose or special-purpose computer. By way of example and not limitation, non-transitory computer-readable media may include RAM, ROM, electrically erasable programmable read-only memory (EEPROM), or optical disk storage devices, magnetic disk storage devices, or other magnetic storage devices, or any other non-transitory media that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a general-purpose or special-purpose computer, or a general-purpose or special-purpose processor.
[0255] Also, any connection is properly referred to as a computer-readable medium. For example, if the software is transmitted from a website, server or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio and microwave are included in the definition of medium. Combinations of the above are also included within the scope of computer-readable media.
[0256] Other examples and implementations are within the scope of the present disclosure and the appended claims. For example, due to the nature of software, the functions described above may be implemented using software executed by a processor, hardware, firmware, hardwiring, or a combination of any of these. Features that implement the functions may also be physically located at various locations, including distributed so that portions of the functions are implemented at different physical locations.
[0257] Also, as used herein, including in the claims, "or" used in a list of items (e.g., a list of items beginning with a phrase such as "at least one of" or "one or more of") indicates an inclusive list, so that, for example, a list of at least one of A, B, or C means A or B or C or AB or AC or BC or ABC (i.e., A and B and C). Also, as used herein, the phrase "based on" should not be understood as referring to a closed set of conditions. For example, an exemplary step described as "based on condition A" may be based on both condition A and condition B without departing from the scope of the present disclosure. In other words, as used herein, the phrase "based on" should be interpreted similarly to the phrase "based at least in part on".
[0258] It will be appreciated from the foregoing that, although specific examples have been described herein for illustrative purposes, various modifications may be made while still maintaining the scope of the desired technology. The description herein is provided to enable a person skilled in the art to perform or use the present disclosure. Various modifications to the present disclosure will be apparent to a person skilled in the art, and the general principles defined herein may be applied to other variations without departing from the scope of the present disclosure. Therefore, the present disclosure is not limited to the examples and designs described herein, but will be given the broadest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method comprising: Calculating at least one layer of multiplication / accumulation processing MAC units, input data and delayed data to generate a result of another layer of MAC units, the another layer of MAC units being different from the at least one layer of MAC units; providing output data based at least in part on the results; and The output data is transmitted via at least one antenna according to a wireless protocol.
2. The method of claim 1, wherein the output data represents the input data processed according to aspects of the wireless protocol. The method of claim 2 , wherein transmitting the output data comprises transmitting the output data according to the wireless protocol.
4. The method according to claim 1, wherein the calculating further comprises: providing a delayed version of the processed input data as input to the at least one layer MAC unit; and The processed delayed version of the input data is multiplied with the input data to generate a multiplication processing result.
5. The method according to claim 4, wherein the calculating further comprises: storing the multiplication processing result at one or more of a plurality of memory lookup units MLU units; retrieving the multiplication process result and a plurality of coefficients at the one or more MLU units of the plurality of MLU units; and The processing result is generated based on the multiplication processing result and the plurality of coefficients.
6. A device comprising: a plurality of transmission devices configured to transmit sensor data; A receiver configured to receive narrowband transmissions from the plurality of transmission devices and configured to process the narrowband transmissions using a plurality of multiply-accumulate MAC units and a plurality of delay units configured to generate output data based on the narrowband transmissions and delayed versions of the narrowband transmissions.
7. The apparatus of claim 6, wherein the narrowband transmission is a narrowband Internet of Things (IoT) transmission.
8. The apparatus of claim 6, wherein the plurality of transmission devices are configured to transmit the sensor data according to a wireless communication protocol.
9. The apparatus of claim 8, wherein the output data represents the narrowband transmission processed according to aspects of the wireless communication protocol.
10. The apparatus of claim 6, wherein one or more delay cells of the plurality of delay cells are configured to introduce one or more time delays into the narrowband transmission.
11. The apparatus of claim 7, wherein in at least one layer of MAC units, one or more delay units of the plurality of delay units coupled to one or more MAC units are configured to introduce the same time delay.
12. The apparatus of claim 6, wherein processing the narrowband transmission using the plurality of MAC units and the plurality of delay units comprises: multiplying at least a portion of the narrowband transmission with a corresponding portion of a delayed version of the narrowband transmission to generate a corresponding coefficient multiplication result; Accumulate the corresponding coefficient multiplication result with one or more of the other corresponding coefficient multiplication results to provide a corresponding processing result to a corresponding MLU unit of a plurality of memory lookup units MLU units; and At one or more MLUs of the plurality of MLU units, one or more coefficients are applied to the processing results to generate the output data.
13. The apparatus of claim 12, wherein the plurality of MAC units, the plurality of delay units, and the plurality of MLU units form a recursive neural network.
14. An apparatus comprising a non-transitory computer-readable medium encoded with executable instructions that, when executed by at least one processing unit, are configured to cause the apparatus to perform operations comprising: Implemented wireless receiver processing mode; Mixing the input data at the at least one processing unit using a plurality of coefficients, wherein mixing the input data comprises: obtaining the plurality of coefficients based on the implemented wireless receiver processing mode; and calculating the input data and delayed versions of the input data using a portion of the plurality of coefficients to generate a plurality of processing results; and Output data is provided based in part on the plurality of processed results.
15. The apparatus of claim 14, further comprising: A radio frequency (RF) antenna is configured to receive an RF wireless transmission including input data.
16. The apparatus of claim 15, further comprising: A recursive neural network includes a plurality of multiplication / accumulation MAC processing units, a plurality of delay units and a plurality of memory lookup unit MLU units.
17. The apparatus of claim 16, wherein the obtaining is at least a portion from the plurality of memory lookup units, wherein the calculating is at least a portion of the MAC processing unit, and wherein the providing is for a mobile application of the apparatus.
18. The apparatus of claim 14, wherein implementing the wireless receiver processing mode is based on a processing mode selection corresponding to the wireless receiver processing mode.
19. The apparatus of claim 14, further comprising a memory lookup unit (MLU) configured to provide corresponding ones of the plurality of processing results, wherein a portion of the plurality of the MLUs is configured to provide output data based at least in part on corresponding converted signals mixed using the plurality of coefficients.
20. The apparatus of claim 15, wherein at least one of the plurality of delay cells is coupled to a corresponding MAC cell of the portion and is configured to provide the delayed version of input data based at least in part on a corresponding output of at least a portion of a MAC cell.
