Deep Learning-Based Online Adaptation of Digital Predistortion and Power Amplifier Systems

JP7580626B2Active Publication Date: 2024-11-11MITSUBISHI ELECTRIC CORP
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
JP2023554095
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2021-01-11
Filing Date
2021-08-12
Publication Date
2024-11-11
Estimated Expiration
2041-08-12

Smart Images

  • Figure 0007580626000001
    Figure 0007580626000001
  • Figure 0007580626000002
    Figure 0007580626000002
  • Figure 0007580626000003
    Figure 0007580626000003
Patent Text Reader

Abstract

An auto-tuning controller for improving the performance of a power amplifier system is provided. The controller includes an interface having an input terminal and an output terminal, the interface being configured to acquire input signal conditions of a power amplifier (PA), and the controller further includes a digital Doherty amplifier (DDA) controller (module), a DPD controller (module), and a training circuit including a processor and memory for executing and storing a DDA-DPD neural network (NN). The training circuit is configured to sample input signal conditions, select a DPD model from a set of polynomial models for the DPD controller and a set of DDA optimization variables for the DDA controller, and use the optimized DPD model and DDA coefficients, where the optimized DPD model and DDA coefficients are provided by performing offline optimization for the DPD model and the DDA coefficients based on a predetermined optimization method, and the training circuit is further configured to collect the optimized DPD coefficients and optimized DDA optimization variables, generate DDA optimal coefficients and DPD optimal coefficients online using the trained DDA-DPD NN, and update the generated DDA optimal coefficients and DPD optimal coefficients via an output terminal of the interface.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates generally to power amplifier systems, and more particularly to a digital Doherty power amplifier system and a learning-based auto-tuning optimization method for a Digital Pre-Distortion (DPD) system and power amplifier system for enhancing the performance of radio frequency power amplifiers. [Background technology]

[0002] The rapid increase in wireless communication data volume and rate significantly increases the power consumption in wireless transmitters, where power amplifiers (PAs) are the main components for energy consumption. Several advanced techniques, including Envelope Tracking (ET), Doherty Power Amplifier (DPA), and Envelope Elimination and Restoration (EER), have been proposed to improve the Power Added Efficiency (PAE) of PAs. Among these techniques, DPA is very promising because its simple structure enables high average efficiency based on active load modulation.

[0003] Although DPA shows many advantages for improving efficiency, traditional analog DPA still suffers from shortcomings that lead to performance degradation in terms of energy efficiency and operating bandwidth. Traditional DPA designs are based on a single input configuration, including an analog power divider (possibly adjustable), fixed phase alignment, a carrier PA running on class AB mode and a peak PA running on class C mode, and an output power combiner. To improve DPA efficiency, several methods have been investigated, including gate bias adaptation, asymmetric DPA, multi-way DPA, adjustable phase alignment, and adaptive power division ratio.

[0004] To obtain optimal digital predistortion (DPD) and PA performance, designers need to manually tune the circuit operating parameters, and the tuning process is only valid for fixed operating conditions such as input power, frequency, and signal standard. In real scenarios, the optimal control parameters change with changing input and circuit conditions. The compensation circuit components are also complex and difficult to optimize, making the DPA design cumbersome. These are precisely the limitations from a purely analog-based design.

[0005] A more flexible architecture, such as digital DPA (DDPA), is needed to adaptively find optimal control parameters for various circuit conditions and input signals with different bandwidths, modulation formats, power levels and modulation formats. Furthermore, in this invention, we propose to auto-adjust the PA's learning cost function as well as auto-adjust the PA's control parameters as a function of the DPD performance, so that both systems improve their performance in unison. Summary of the Invention

[0006] Some embodiments are based on the recognition that a digital power amplifier (DPA) is programmable so that it can facilitate a circuit tuning procedure (auto-tuning) for a designer and can take into account circuit imbalances such as phase delay for multiple paths and environmental changes including temperature and aging. Thus, a DPA can provide not only flexibility but also improved performance compared to an analog DPA.

[0007] Some embodiments of the present invention also provide an auto-tuning controller for improving the performance of a power amplifier system. The auto-tuning controller may include an interface including an input terminal and an output terminal. The interface is configured to obtain an input signal condition of a power amplifier (PA). The auto-tuning controller may further include a digital Doherty amplifier (DDA) controller (module), a DPD controller (module), and a training circuit including a processor and a memory for executing and storing a DDA-DPD neural network (NN). In this case, the training circuit is configured to sample input signal conditions, select a DPD model from a set of polynomial models for the DPD controller and a set of DDA optimization variables for the DDA controller, and use the optimized DPD model and DDA coefficients, where the optimized DPD model and DDA coefficients are provided by performing offline optimization for the DPD model and the DDA coefficients based on a predetermined optimization method, and the training circuit is further configured to collect the optimized DPD coefficients and the optimized DDA optimization variables, generate DDA optimal coefficients and DPD optimal coefficients online using the trained DDA-DPD NN, and update the generated DDA optimal coefficients and DPD optimal coefficients via an output terminal of the interface.

[0008] According to an embodiment of the present invention, a DPA system, a Digital Doherty Power Amplifier (DDPA) system, a digital predistortion (DPD), and a learning-based automatic adjustment method (optimization method) are provided, which notably improve the efficiency and gain while the DPD is simultaneously working with the PA system by adaptive control to meet linearity requirements. The DDPA system and optimization method can be used for wideband mobile communications including 3G, 4G LTE, and 5G base station radio front-ends of transmitters.

[0009] Further, according to some embodiments of the present invention, a digital Doherty power amplifier (DDPA) system is provided. The DDPA may include an automatic tuning controller for improving the performance of the power amplifier system. In this case, the automatic tuning controller may include an interface including an input terminal and an output terminal. The interface is configured to obtain an input signal condition of the power amplifier (PA). The automatic tuning controller may further include a digital Doherty amplifier (DDA) controller (module), a DPD controller (module), and a training circuit including a processor and a memory for executing and storing a DDA-DPD neural network (NN). In this case, the training circuit is configured to sample the input signal conditions to select a DPD model from a set of polynomial models for the DPD controller and a set of DDA optimization variables for the DDA controller, and to use the optimized DPD model and DDA coefficients, where the optimized DPD model and DDA coefficients are provided by performing offline optimization for the DPD model and the DDA coefficients based on a predetermined optimization method, and the training circuit is further configured to collect the optimized DPD coefficients and the optimized DDA optimization variables, generate DDA optimal coefficients and DPD optimal coefficients online using the trained DDA-DPD NN, and update the generated DDA optimal coefficients and DPD optimal coefficients via an output terminal of the interface, and the DDPA system may further include a Doherty power amplifier (DPA) circuit having a control input and an output for generating an output signal.

[0010] Some embodiments of the present invention provide a DPD and digital Doherty amplifier (DDA) system that is fully adaptive to find an optimal control parameter set without complex engineering tuning despite device parameter and environmental variations, and the optimal control aims for high efficiency and reasonable gain in, for example, a wideband power transmitter.

[0011] One embodiment of the invention is a model-free algorithm that has no assumptions or prior knowledge about the DPA device, and the algorithm is based on black-box optimization to search for an optimal configuration. The optimal configuration consists of the optimal coefficients of the DPD model and the PA optimal coefficients. This optimal configuration is obtained offline for different operating conditions such as input power conditions, frequency bandwidth, and signal modulation. When enough optimal data is collected offline, we propose to train a deep neural network (DNN) model that maps the operating conditions to the optimal DPD and / or DPA configurations.

[0012] Some embodiments not only optimize for DPA efficiency, but also flexibly improve gain and linearity performance, while the linearity characteristic term in the learning cost is proportional to the DPD linearization performance. For example, in some embodiments, the system can balance the tradeoff between gain and efficiency in different frequency bands, or maximize efficiency under certain constraints. One example is optimizing efficiency while requiring a gain larger than a configured threshold. For modulated signals, we optimize efficiency, gain, along with Adjacent Channel Power Ratio (ACPR) under the same DDPA scenario. ACPR is an important linearity indicator for modulated signals to have limited power emitted from the main channel to adjacent channels.

[0013] According to an embodiment of the present invention, a digital power amplifier (DPA) system includes a power amplifier (PA) circuit having a control input and an output for generating an output signal, and an adaptive control circuit, the adaptive control circuit comprising an input interface, an output interface, a memory for storing an adaptive control algorithm, and a processor for performing instructions based on the adaptive control algorithm associated with the memory, the input interface receiving an input state signal and an output signal of the PA circuit, and the adaptive control algorithm, in response to the input state signal and the output signal, determines a control parameter of a control signal sent from the output interface to the control input for controlling operation of the PA circuit.

[0014] The accompanying drawings, which are included to provide a further understanding of the invention, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention. [Brief description of the drawings]

[0015] [Figure 1] 1 is a block diagram illustrating a prior art digital power amplifier (DPA). [Diagram 2] 1 is a schematic diagram illustrating stages in improving the linearity and efficiency of a power amplifier through a DPD (digital pre-distortion) process, according to the related art; [Figure 3A] FIG. 1 is a schematic diagram illustrating a detailed block diagram of a new algorithm for PA auto-adjustment according to the related art. [Figure 3B] 1 is a schematic diagram illustrating a block diagram of a digital power amplifier (DPA) in accordance with an embodiment of the present invention. [Figure 4A] FIG. 1 is a schematic diagram illustrating a block diagram showing a shared control module for DPD and DDA, according to an embodiment of the present invention. [Figure 4B] FIG. 1 is a schematic diagram illustrating a block diagram showing an implementation of a shared control module for DPD and DDA using deep neural networks according to an embodiment of the present invention. [Figure 4B-1]FIG. 1 is a block diagram illustrating an implementation of a DNN shared control module for DPD and DDA in accordance with an embodiment of the present invention. [Figure 4B-2] FIG. 2 is a block diagram illustrating an implementation of a DPD model and DDA optimization coefficients according to an embodiment of the present invention. [Figure 4C] FIG. 2 is a block diagram illustrating an implementation of a shared control module for DPD and DDA using deep neural networks, according to an embodiment of the present invention. [Figure 4C-1] FIG. 1 is a block diagram illustrating an implementation of a DNN shared control module for DPD and DDA in accordance with an embodiment of the present invention. [Figure 4D] FIG. 2 is a block diagram illustrating implementation of input signal conditioning according to an embodiment of the present invention. [Figure 4E] FIG. 2 is a block diagram illustrating several options for DNNs for the DDA DNN control module and the DPD DNN control module, according to an embodiment of the present invention. [Figure 4F] FIG. 11 is a block diagram illustrating alternative DNN options for the DDA DNN control module and the DPD DNN control module, in accordance with an embodiment of the present invention. [Diagram 5] FIG. 2 is a block diagram illustrating a transfer learning process for a DPD control module and a DDA control module in accordance with an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] Various embodiments of the present invention are described below with reference to the drawings. It should be noted that the drawings are not drawn to scale, and elements of similar structure or function are represented by the same reference numerals throughout the drawings. Furthermore, the drawings are merely intended to facilitate the description of specific embodiments of the present invention. The drawings are not intended as an exhaustive description of the present invention or as limitations on the scope of the present invention. In addition, aspects described in connection with a particular embodiment of the present invention are not necessarily limited to that embodiment, and may be practiced in any other embodiment of the present invention.

[0017] FIG. 1 is a block diagram of a digital power amplifier (DPA) module 100 according to some embodiments of the present invention. The DPA module 100 may be a digital power amplifier (DPA) module configured by a multi-input power amplifier 120, such as a Doherty power amplifier, an outphasing power amplifier, a balanced power amplifier, and a push-pull power amplifier. As an example, a Doherty power amplifier is used in the digital power amplifier (DPA) module 100 to explain the function of the DPA module. The DPA module 100 may be called a DDPA (digital Doherty power amplifier) ​​module 100. However, depending on the variation of the circuit design, an outphasing power amplifier circuit, a balanced power amplifier circuit, or a push-pull power amplifier circuit may also be used.

[0018] The DDPA module 100 may include a baseband processing module 101, a digital Doherty amplifier (DDA) adaptive control module 102, an amplitude ratio-phase control module (amplitude (Amp)-phase module) 103, a signal converter 110, a dual-input DPA module (DPA module, but not limited to dual-input) 120, and a power supply 104 for providing bias conditions (voltage and / or current) to the DPA module 120 according to optimal control parameters (or control parameters) 313 generated by the DDA adaptive control module 102. The DPA module 120 includes a main PA (carrier PA) 121, a peak PA 122, and an output coupler 123. The signal converter 110 includes digital-to-analog converters (DACs) 111 and 112, and up-converters 113 and 114.

[0019] The dual-input DPA module 120 includes a carrier power amplifier (PA) 121 for controlling the carrier signal, a peak power amplifier (PA) 122 for controlling the peak signal, and an output combiner 123 for combining signals from the carrier PA 121 and the peak PA 122. It is apparent that the digital Doherty configuration can be extended to a multi-way Doherty in which more than two PAs are associated with a similar topology described in 100.

[0020] If a power amplifier other than a Doherty amplifier is used in the module 100, the DDA adaptive control module 102 may be referred to as a digital adaptive (DA) control module 102.

[0021] The input signal 301 is sent to the DA adaptation control module 102. The DA adaptation control module 102 uses the input signal 301 and the output signal 300 from the DPA module 120 to perform data-driven optimization to generate optimal control parameters 313 for the phase difference θ between the PA 121 and the PA 122 and the input power ratio α for the PA 121 and the PA 122. In some cases, the control parameters 313 may be referred to as updated DDPA parameters. The updated DDPA parameters 313 are provided to the amplitude-phase module 103. The above-described optimization may also be referred to as a learning-based auto-tuning method for the power amplifier system.

[0022] In this case, the control parameters 313 include gate bias parameters (Vg1, Vg2) of PA121 and PA122, a phase difference θ between PA121 and PA122, and an input power distribution (ratio α) for PA121 and PA122. The power supplier 104 also receives the gate bias parameters from the DDA adaptive control module 102 and applies gate bias voltages to PA121 and PA122 according to the gate bias parameters (Vg1, Vg2). When the amplitude-phase module 103 receives the phase difference between PA121 and PA122 and the input power ratio for PA121 and PA122 from the DDA adaptive control module 102 as part of the control parameters 313, the amplitude-phase module 103 generates signals S1 and S2 to be applied to PA121 and PA122, respectively. In this case, the signals S1 and S2 are formed such that the amplitude ratio of the signals S1 and S2 and the phase difference between the signals S1 and S2 satisfy the values ​​indicated by the optimal control parameters 313 calculated by the DDA adaptive control module 102.

[0023] In some cases, the DDA adaptive control module 102 may include a power supply 104, and the PA 121 and the PA 122 may be field-effect-transistors (FETs) fabricated using gallium nitride (GaN)-based materials or other different semiconductor device technologies. The PA 121 and the PA 122 may also be bipolar transistors (BPTs). In this case, the gate bias is replaced by the base current bias of the bipolar transistor. The BPTs may be formed by GaN-based materials or other different semiconductor device technologies.

[0024] The output signal of the output coupler 123 (for example, a quarter-wave transmission line can be used as the output coupling network, and other formats such as capacitor-based, lumped element inductors, etc. can also be used to perform the same coupling function) can be sent from an antenna (not shown) through a predetermined bandpass filter (not shown). The output signal of the output coupler 123 is also received (detected) by an input interface (not shown) of the DDA control module 102 to calculate control parameters 313 that control the carrier PA 121 and the peak PA 122, respectively. In some cases, the control parameters 313 can be referred to as tuning parameters.

[0025] Each of the carrier PA 121 and the peak PA 122 may be controlled by control parameters 313 indicating gate bias, phase, and input signal power. The phase signal and the input power signal are applied to the amplitude-phase module 103. The amplitude-phase module 103 adjusts the amplitude ratio and phase of the carrier PA 121 and the peak PA 122 signals according to the control parameters 313 generated by the DDA adaptation control module 102.

[0026] In this case, the signal converter 110 generates and provides phase and input power signals to the carrier PA 121 and the peak PA 122 via drive amplifiers (not shown) that drive the carrier PA 121 and the peak PA 122, if necessary.

[0027] The DDA adaptive control module 102 includes a processor (not shown) associated with one or more memories (not shown) that store adaptive control algorithms, the processor executing instructions according to a predefined adaptive control algorithm, and the adaptive control algorithm is based on model-less optimization called adaptive tuning control.

[0028] The control parameters 313 are generated by the DDA adaptive control module 102, which calculates values ​​for the control parameters 313 through adaptive optimization control. In some cases, the control parameters 313 may be referred to as data-driven optimization parameters.

[0029] Furthermore, the DDA adaptation control module 102 also includes an input interface (not shown) for receiving the input signal 301 and the output signal 300 of the dual input DPA module 120, and an output interface (not shown) for generating control parameters 313 including phase control signals, power ratio control signals, and bias signals for controlling the dual input DPA module 120 and the carrier PA 121 and the peak PA 122.

[0030] The DDA adaptive control module 102 receives a baseband signal from the baseband processing module 101, detects the output signal 300 of the output combiner 123 through an input interface, and generates control parameters 313, which may be called data-driven optimization parameters 313, based on an adaptive control algorithm. In this case, a part of the data-driven optimization parameters 313 is provided to the amplitude-phase control module 103 through an output interface to control the phase and power ratio of the PA121 and the PA122. And another part of the data-driven optimization parameters 313 is converted into gate biases that are respectively provided to the PA121 and the PA122 through the power supply 104 to control the gate biases of the PA121 and the PA122.

[0031] In Fig. 1, the DDPA module 100 is programmable so that it can fully consider the circuit imbalance and shortcomings for multiple paths, facilitating the circuit tuning procedure for the designer. Therefore, the DDPA module 100 provides not only flexibility and low cost but also better performance compared to analog DPA. The design according to an embodiment of the present invention benefits from the principle of software design, so that the control port can be adapted to reach optimal performance through an algorithm.

[0032] Making an RF power amplifier more efficient means driving it to a point near its saturation point. In such cases, the modulated waveform is prone to distortion (incurring nonlinearities characterized by ACPR (adjacent channel power ratio)). Thus, the design goal is to maximize power-added efficiency (PAE) while maintaining high gain and good linearity (ACPR). In some cases, digital linearity can be achieved through digital predistortion (DPD).

[0033] 2 shows the stages in improving the linearity and efficiency of PA121 and PA122, respectively, through a DPD (digital pre-distortion) process 201 and an efficiency enhancement process 202. In these processes, digital pre-distortion of an input signal is performed in step 201, efficiency enhancement is performed in step 202, and the input signal obtained by the process through digital pre-distortion (DPD) and efficiency enhancement is provided to PA121 and PA122 in step 203.

[0034] 3A shows a detailed block diagram of our algorithm for PA auto-adjustment, in which an input signal (input state signal) 301 is used as an input state including different states. The different states indicate frequency and input power level, but are not limited to these two states, and can include other states such as signal modulation format. The input state signal 301 indicating frequency and input signal power level is fed into the DDA adaptive control module 102, which adaptively adjusts the control parameters 313 of the DDPA module 120, such as gate bias 303, phase difference between the main amplifier and the peak amplifier 304, and input power distribution 305, by using the input signal 301 and the output signal 300 of the dual input DPA module 120. Finally, the adjusted parameters are fed into the dual input DPA module 120.

[0035] According to variations in circuit design, the DPA module 120 may include three or more power amplifiers (PAs). For example, see FIG. 3B. In such a case, the DDA adaptive control module 102 provides control parameters 313 for each of the three or more power amplifiers.

[0036] FIG. 3B is a block diagram illustrating a multi-input digital power amplifier module (eg, a digital Doherty power amplifier (DDPA) system) 350 that performs an automatic tuning process of the DPA module in accordance with an embodiment of the present invention.

[0037] In the figure, if the function of a part is the same as that in Fig. 1, the same part number as in Fig. 1 is used for the part, and the description of the same part number is omitted.

[0038] The multi-input digital power amplifier module 350 includes a baseband processing module 101, a DDA adaptive control module (automatic tuning controller) 351, an amplitude ratio / phase control module 103, a signal converter 110, and a multi-input PA module 120. In this case, the signal converter 110 includes digital-to-analog converters (DACs) 111, 112, and 112n, and up-converters 113, 114, and 114n.

[0039] The DDA adaptation control module 351 may be referred to as a shared control module for DPD and DDA. For example, the configuration or function is described below with reference to FIGS.

[0040] The DDA adaptive control module 351 includes an adaptive controller 352, a deep learning based online DPD 353, and components not shown in the drawing. For example, the auto-tuning controller 351 includes a training circuit and an interface including an input terminal and an output terminal. The interface is configured to obtain a signal 301 from the baseband processing module 101, obtain a signal condition 300 from the power amplifier (PA) 120, and transmit an optimal control parameter (optimal control parameter signal) 313.

[0041] The training circuit includes a processor and memory for executing and storing a digital Doherty amplifier (DDA) controller (module), a DPD controller (module), and a DDA-DPD neural network (NN). The training circuit is configured to sample an input signal condition to select a DPD model from a set of polynomial models for the DPD controller and a set of DDA optimization variables for the DDA controller, and to use an optimized DPD model and DDA coefficients. In this case, the optimized DPD model and DDA coefficients are provided by performing offline optimization for the DPD model and DDA coefficients based on a predetermined optimization method. The training circuit is further configured to collect the optimized DPD coefficients and the optimized DDA optimization variables, generate DDA optimal coefficients and DPD optimal coefficients online using the trained DDA-DPD NN, and update the generated DDA optimal coefficients and DPD optimal coefficients via an output terminal of the interface.

[0042] The multi-input PA module 120 includes a carrier power amplifier (PA) 121 for controlling a carrier signal, a peak power amplifier (PA) 122 for controlling a peak signal, a second peak power amplifier (PA) 122n for controlling a second peak signal, and an output combiner 123 for combining signals from the PAs 121, 122, and 122n. In this case, the signal converter 110 includes three or more DACs and three or more up-converters, and the multi-input PA module 120 includes three or more power amplifiers 121, 122, and 122n.

[0043] The DDA adaptation control module 351 uses the input signal 301 and the output signal 300 of the multi-input PA module 120 to generate updated DPA parameters 313 and provides the updated DPA parameters 313 to the amplitude-phase module 103. The amplitude-phase module 103 then provides a signal to the signal converter 110 for signal conversion to generate S1, S2 and Sn to be applied to PA121, PA122 and PA122n, respectively.

[0044] As mentioned above, the optimal control parameters 313 are calculated using the adaptive control module 351. A detailed description of the optimal control parameters is provided below.

[0045] In Fig. 4A, one embodiment 450 of the present invention is a design of a shared control module 410 for DPD and DDA, which sends optimal coefficients to the DDA control module 405 and the DPD control module 401. In some embodiments of the present invention, we propose that the shared control module 410 is realized as follows: In Fig. 4B, for a given input signal condition 4100, a DDA-DPD deep neural network (DNN) 4101 generates online the DDA optimal coefficients 4103 that will be sent to the DDA and the DPD optimal coefficients 4102 that will be sent to the DPD model.

[0046] As in FIG. 4B-1, the DDA-DPD DNN 4101 needs to be trained offline (41014). The training process follows the following steps. First, the input signal conditions 4100 are sampled from different frequencies 41001, power signals 41002, signal modulation formats 41003, etc. 41004 as in FIG. 4D, then a model of DPD is selected and a set of optimization variables for DDA is selected (41011). As in FIG. 4B-2, a DPD model can be selected (410112), in one embodiment, the DDA coefficients are selected from a set of polynomial models with different sizes 410114, in another embodiment, a nonlinear trigonometric model or other nonlinearity is selected (410115), and in yet another embodiment, a deep neural network model is selected as the DPD model (410116). A set of optimization variables for the DDA is selected 41011 from coefficients such as PAPR (peak to average power ratio) reduction threshold 410120, power ratio 410121, phase difference 410117, attenuation difference, power amplifier main bias voltage 410118, power amplifier peak bias voltage 410119, etc. This optimization of DPD and DDA can be done using extreme value search optimization 41015, simulated annealing 41016, Bayesian optimization 41017, hill climbing 41018, least squares 41019, etc. 41029. Once this optimization is done, we collect the DDA and DPD optimal coefficients 41013, which are then used along with the input signal condition 4100 to train the DDA-DPD DNN 41014 offline. In this training stage, the input of the DDA-DPD DNN 41014 is the input signal condition 4100, and the output of the DDA-DPD DNN 41014 is the DDA and DPD optimal coefficients 41013. The training of this DDN can be done by any DDN training method. For example, in some embodiments, we can use Adam's method, or gradient descent method, or stochastic gradient descent method, etc.If the training of the DDN 41014 converges, we obtain a DDA-DDP DNN to be used online to generate DDA and DPD optimal coefficients for any new input signal condition 4100.

[0047] In Fig. 4C, as another embodiment, the shared control module 410 can be realized as follows: For a given input signal condition 4100, two DDNs are obtained: DDA DNN 4121 and DPD DNN 4131. The DDA DNN 4121 generates DDA optimal coefficients 4103 online, which will be sent to the DDA. The DPD DNN 4131 generates DPD optimal coefficients 4102 online, which will be sent to the DPD.

[0048] As in Figure 4C-1, two separate DDA DNNs 4102 and DPD DNNs 4103 need to be trained offline (41211, 41311). The training process follows the following steps:

[0049] First, the input signal conditions are sampled from different frequencies, power signals, signal modulation formats, etc. (410111), then a model of DPD is selected and a set of optimization variables for DDA is selected (41011). In one embodiment, the DPD model can be selected from a set of polynomial models with different sizes 410114 (410122) (410112), in another embodiment, a nonlinear trigonometric model or other nonlinearity is selected (410115), and in yet another embodiment, a deep neural network model is selected as the DPD model (410116). A set of optimization variables for DDA is selected from coefficients such as PAPR reduction threshold 410120, power ratio 410121, phase difference 410117, attenuation difference, power amplifier main bias voltage 410118, power amplifier peak bias voltage 410119, etc. (410113). Then, offline optimization of DDA and DPD is performed (41012). This optimization of DPD and DDA can be done using extreme value search optimization 41015, simulated annealing 41016, Bayesian optimization 41017, hill climbing 41018, least squares 41019, etc. 41029. Once this optimization is done, we collect DDA and DPD optimal coefficients 41013, which are then used with input signal condition 4100 to train DDA DNN 41211, and DPD DNN 41311 offline. For training DDA DNN, the DNN input is input signal condition 4100, and the output of DDA DNN 41211 is DDA optimal coefficients 4103. For training DPD DNN, the DNN input is input signal condition 410111, and the output of DPD DNN 41311 is DPD optimal coefficients 4102. Training of these DDNs can be done by any DDN training method. For example, in some embodiments, we can use Adam's method, or gradient descent, or stochastic gradient descent, robust DNN training, etc. When the training of the DDA DDN 41211 converges, we obtain a DDA DNN to be used online to generate DDA optimal coefficients for any new input signal condition 410111.If the training of the DPD DDN 41311 converges, we obtain a DPD DNN to be used online to generate DPD optimal coefficients for any new input signal condition 4100.

[0050] As in Figures 4E-4F, in different embodiments of the present invention, we can use different types of DNNs 4101, 4121, 4131. For example, fully connected DNN 41011, convolutional neural network (CNN) 41012, residual network (ResNet) 41013, recurrent neural network (RNN) 41014, etc. 41015.

[0051] FIG. 5 is a block diagram illustrating a transfer learning process for a DPD control module and a DDA control module in accordance with an embodiment of the present invention.

[0052] In some embodiments of the present invention, a transfer learning procedure is performed from one unit 500 to another unit 501, 5100. For example, in one embodiment of such an implementation of transfer learning, we propose to start with a given unit 1 500, and then for a given input condition 4100, we calculate the corresponding DDA and DPD optimal coefficients 41013 offline, which are then used to train 41014 the corresponding DNN associated with unit 1 500 offline. Then, for the next unit 2 501, for a given input condition 4100 that is similar or different to that of unit 1, we calculate the DDA and DPD optimal coefficients 41013 offline, which are then used for offline training 521 of the DNN corresponding to unit 2. The trick for transfer learning is to use the optimal coefficient Opt_coeff_1 for the DNN for unit 1 to warm start the training 510 of the DNN for unit 2. Next, for another unit N 5100, we proceed with similar training and we start training a new DNN using the optimal coefficients of the DNN for unit N-1 5100.

[0053] This transfer learning between units can also be embodied by fixing some of the DNN coefficients to their optimal values ​​obtained for the previous unit and retraining the new DNN associated with the new unit by training only some of the DNN's coefficients.

[0054] Yet another transfer learning embodiment can be performed by extending the DNN with additional layers for each new unit and retraining the DNN while learning only the coefficients of the additional layers and keeping the coefficients of the other layers equal to the optimal coefficients obtained for the previous unit.

[0055] The above-described embodiments of the present invention may be implemented in any of many ways. For example, the embodiments may be implemented using hardware, software, or a combination thereof. When implemented in software, the software code may be executed on any suitable processor or collection of processors, whether the processors are provided in a single computer or distributed among multiple computers. Such a processor may be implemented as an integrated circuit having one or more processors within an integrated circuit component. However, the processor may be implemented using circuits in any suitable format.

[0056] Also, embodiments of the invention may be embodied as a method, examples of which are provided. The operations performed as part of the method may be ordered in any suitable manner. Thus, while an example embodiment may show operations as sequential, embodiments may be constructed in which operations are performed in an order different from that illustrated, including even including performing some operations simultaneously.

[0057] The use of ordinal terms such as "first," "second," etc. in the claims to modify claim elements does not, by itself, imply any priority, precedence, or ordering of a claim element over another claim element, or any chronological order in which method operations are performed, but is merely used as a label to distinguish a claim element having a certain name from another element having the same name (except for the use of the ordinal terminology).

[0058] Although the invention has been described by way of examples of preferred embodiments, it is to be understood that various other adaptations and modifications can be made within the spirit and scope of the invention.

[0059] Therefore, it is the object of the appended claims to cover all such variations and modifications as come within the true spirit and scope of the invention.

Claims

1. 1. An auto-tuning controller for improving performance of a digital Doherty power amplifier (DDPA) system, comprising: an interface including an input terminal and an output terminal, the interface being configured to acquire an input signal condition of a multi-input power amplifier (PA) included in a multi-input PA module of the DDPA system; and the automatic tuning control device further includes: a digital Doherty amplifier (DDA) controller, a digital predistortion (DPD) controller, and a training circuit including a processor and memory for executing and storing a DDA-DPD neural network (NN), said training circuit comprising: sampling the input signal conditions to select a DPD model from a set of polynomial models for the DPD controller and to select a set of DDA optimization variables for the DDA controller; collecting optimized DPD coefficients and optimized DDA optimization variables by performing optimization for the DPD model and the DDA optimization variables based on a predetermined optimization method; generating DDA optimal coefficients and DPD optimal coefficients using a trained DDA-DPD NN, the trained NN being trained offline based on both the input signal conditions and the optimized DPD coefficients and the optimized DDA optimal variables; and updating the generated DDA optimum coefficients and DPD optimum coefficients via the output terminal of the interface; The DDA optimization variables include a peak-to-average power ratio reduction (PAPR) threshold, a power ratio, a phase difference, an attenuation difference, a main bias voltage of the PA, and a peak bias voltage of the PA; Offline training is done using Adam's algorithm, gradient descent, or stochastic gradient descent. the input signal conditions include at least a frequency of the PA, a signal power level, and a signal modulation format; The auto-tuning control device, wherein the predetermined optimization method is an extreme value searching optimization, a simulated annealing, a Bayesian optimization, a hill climbing method, a genetic evolution method, or a least squares method.

2. The auto-tuning controller of claim 1 , wherein the offline training is performed based on a previously optimized DPD model and DDA coefficients.

3. The auto-tuning control device of claim 1 , wherein the DDA-DPD neural network (NN) is a convolutional neural network (CNN), a deep neural network (DNN), or a robust DNN.

4. The auto-tuning controller of claim 1 , wherein the interface is configured to connect to at least two power amplifiers in the multi-input PA module.

5. The automatic adjustment control device according to claim 1 ; and a Doherty power amplifier (DPA) circuit having a control input and an output for generating an output signal.

6. 6. The DDPA system of claim 5, wherein the DDA optimization variables include a peak-to-average power ratio reduction (PAPR) threshold, a power ratio, a phase difference, an attenuation difference, a main bias voltage of the PA, and a peak bias voltage of the PA.

7. The DDPA system of claim 5 , wherein the offline training is performed by Adam's method, gradient descent, or stochastic gradient descent.

8. The DDPA system of claim 5, wherein the DDA-DPD neural network (NN) is a convolutional neural network (CNN), a deep neural network (DNN), or a robust DNN.

9. The DDPA system of claim 5, wherein the trained DDA-DPD NN is trained offline based on both the input signal conditions and the optimized DPD and DDA coefficients.

10. 6. The DDPA system of claim 5, wherein the predetermined optimization method is extremum seeking optimization, simulated annealing, Bayesian optimization, hill climbing, genetic evolution, or least squares.

Citation Information

Patent Citations

  • Amplification device

    JP2013074595A

  • Transmitter and radio base station

    JP2013239880A

  • Amplifier and communication device

    JP2018142798A

  • Digital Power Amplifier (DPA) System and Digital Doherty Power Amplifier (DDPA) System

    JP2020526995A

  • Predistortion Control Loop for RF Power Amplifiers

    JP2020537847A