Method and apparatus for machine learning based radio frequency (RF) front-end calibration

Abstract

Some aspects of the present disclosure provide techniques and apparatus for calibrating a radio frequency (RF) circuit using machine learning. One exemplary method generally includes calibrating a first subset of RF circuit calibration parameters. Based on the machine learning model and the first subset of the RF circuit calibration parameters, values ​​are predicted for a second subset of the RF circuit calibration parameters. The second subset of the RF circuit calibration parameters may be different from the first subset of the RF circuit calibration parameters. At least the first subset of the RF circuit calibration parameters is verified, and after verification, at least the first subset of the RF circuit calibration parameters is written to a memory associated with the RF circuit.

Description

[Technical field]

[0001]

[0001] This patent application claims priority to U.S. patent application Ser. No. 17 / 650,334, filed Feb. 8, 2022, the entire contents of which are expressly incorporated by reference into this specification.

[0002] Introduction Aspects of the present disclosure relate to radio frequency (RF) circuit calibration.

[0003]

[0003] Radio frequency circuits generally allow signaling to be converted to and from radio frequency bandwidths for transmission to or processing of signaling received from other devices. These RF circuits are generally manufactured as a set of electronic circuits forming a complete unit. However, due to variations in the manufacturing process, the characteristics of each RF circuit may vary. Therefore, to enable these RF circuits to operate at a similar level, each RF circuit may be individually calibrated, and the parameters with which each RF circuit is calibrated may be written into a memory associated with the RF circuit.

[0004]

[0004] The complexity involved in calibrating RF circuits generally increases as these RF circuits are designed to support new functions. For example, RF circuits may be calibrated for operation in various frequency bands (e.g., FR1 band between 4.1 GHz and 7.125 GHz, FR2 band between 24.25 GHz and 52.6 GHz, etc.), for operation using a single antenna, or for operation using multiple antennas in a multiple-input-multiple-output (MIMO) system, etc. For RF systems that implement MIMO technology or enable communication using millimeter wave frequencies, the number of uplinks and downlinks in the system may further increase the number of parameters to be calibrated. Thus, as RF circuits become more complex and as the number of configuration parameters for these RF circuits increases, the process of calibrating RF circuits may become increasingly resource intensive.

[0005]

[0005] Therefore, there is a need for techniques to efficiently calibrate parameters of RF circuits. Summary of the Invention

[0006]

[0006] Some aspects provide a method for calibrating a radio frequency (RF) circuit. The method generally includes calibrating a first subset of RF circuit calibration parameters. Based on the machine learning model and the first subset of the RF circuit calibration parameters, values ​​are predicted for a second subset of the RF circuit calibration parameters. The second subset of the RF circuit calibration parameters may be different from the first subset of the RF circuit calibration parameters. At least the first subset of the RF circuit calibration parameters is verified, and after verification, at least the first subset of the RF circuit calibration parameters is written to a memory associated with the RF circuit.

[0007]

[0007] Another aspect provides a processing system configured to perform the aforementioned methods as well as methods described herein; a non-transitory computer readable medium including instructions that, when executed by one or more processors of the processing system, cause the processing system to perform the aforementioned methods as well as methods described herein; a computer program product embodied on the computer readable storage medium including code for performing the aforementioned methods as well as methods further described herein; and an apparatus comprising means for performing the aforementioned methods as well as methods further described herein.

[0008] The following description and the annexed drawings set forth in detail certain illustrative features of the one or more embodiments. [Brief description of the drawings]

[0009]

[0009] The accompanying drawings illustrate several aspects of one or more embodiments and therefore should not be considered as limiting the scope of the present disclosure. [Figure 1]

[0010] 1 illustrates an example pipeline for calibrating a radio frequency (RF) circuit. [Figure 2A]

[0011] 1 illustrates an example pipeline for off-chip calibration of RF circuits using machine learning models, according to an aspect of the present disclosure. [Figure 2B] 1 illustrates an example pipeline for off-chip calibration of RF circuits using machine learning models, according to an aspect of the present disclosure. [Figure 3A]

[0012] 1 illustrates an example pipeline for on-chip calibration of RF circuits using machine learning models, according to an aspect of the present disclosure. [Figure 3B] 1 illustrates an example pipeline for on-chip calibration of RF circuits using machine learning models, according to an aspect of the present disclosure. [Figure 4]

[0013] 1 illustrates an example pipeline for real-time calibration of RF circuits using machine learning models, according to an embodiment of the present disclosure. [Diagram 5]

[0014] 1 illustrates example operations that may be performed by an RF circuit calibration system to calibrate an RF circuit using a machine learning model, according to aspects of the present disclosure. [Figure 6]

[0015] 1 illustrates an example implementation of a processing system capable of calibrating RF circuits using machine learning models in accordance with aspects of the present disclosure.

[0010]

[0016] For ease of understanding, wherever possible, the same reference numbers have been used to designate identical elements common to the figures. It is contemplated that elements and features of one embodiment may be advantageously incorporated in other embodiments without further recitation. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011]

[0017] Aspects of the present disclosure provide apparatus, methods, processing systems, and computer-readable media for calibrating radio frequency (RF) circuits using machine learning techniques.

[0012]

[0018] RF circuits are subject to variations in manufacturing and operating parameters that may affect the operation of such integrated circuits. For example, variations in manufacturing, such as variations in etch depth, thickness of metal or oxide layers, impurity concentrations, etc., may affect the resistance, threshold voltage, etc. of individual circuits. Furthermore, as RF circuits become more complex, additional parameters may be configured for each RF circuit. For example, with the introduction of additional functionality to support MIMO technology, higher frequency bands (e.g., FR1 or FR2), etc., new parameters for calibration may be introduced to account for the increased number of uplinks and downlinks in the RF circuit, and more generally, to account for additional components in the RF circuit to implement support for this additional functionality.

[0013]

[0019] Aspects of the present disclosure provide techniques for calibrating an RF circuit using at least some parameters predicted using machine learning techniques. In general, the predicted parameters may be a subset of a population of parameters that may configure the RF circuit and may be predicted based on a set of calibrated parameters of the RF circuit. By calibrating a first set of RF circuit calibration parameters and using the first set to predict a second set of RF circuit calibration parameters (e.g., using machine learning), the computing resources used in calibrating the RF circuit may be reduced relative to calibrating the entire population of RF circuit calibration parameters for any given RF circuit.

[0014] Exemplary Radio Frequency (RF) Circuit Calibration Pipeline

[0020] 1 illustrates an exemplary pipeline 100 in which an RF circuit is calibrated. As shown, pipeline 100 includes an RF parameter calibrator 110, a temporary memory 120, an automatic test engine 125, a code verifier 130, and a field programmable read-only memory (FPROM) 140.

[0015]

[0021] The RF circuit to be calibrated generally includes several circuits that enable upconversion of baseband or intermediate frequency signals to radio frequency signals for transmission, and downconversion of received radio frequency signals to intermediate or baseband frequency signals for processing. These circuits may include, for example, power amplifiers, low noise amplifiers, mixers, filters, frequency synthesizers, phase-locked loops, etc. In general, the number of components in an RF circuit may increase as the complexity of the RF circuit increases, for example, to support additional frequency bands (e.g., for carrier aggregation), a larger number of antennas (e.g., diversity reception), etc. As the number of components in an RF circuit increases, the number of circuit parameters to be calibrated may also increase.

[0016]

[0022] For a number N of parameters to be calibrated, RF parameter calibrator 110 can test and calibrate each parameter and generate N calibration codes for verification. These calibration codes can include, for example, code to adjust power usage parameters, control current bias in the RF circuitry, measure and control voltages at the RF circuitry, etc. In general, as the total number of circuit parameters increases, the total number of calibrations performed by RF parameter calibrator 110 can increase.

[0017]

[0023] The N calibration codes may be written to the temporary memory 120, and the automatic test engine 125 may access the N calibration codes stored in the temporary memory to evaluate the RF circuit being calibrated (e.g., a first pass through the measurement system). In general, the automatic test engine 125 may perform various measurements of the current draw, voltage, etc., of the RF circuit being calibrated, and may determine whether additional calibrations should be performed. Additionally, the automatic test engine 125 may record the calibration codes generated by the RF parameter calibrator 110 for future use.

[0018]

[0024] The N calibration codes may also be provided to the code verifier 130 for verification. In some aspects, the N calibration codes may be provided to the code verifier 130 for verification based on instructions generated by the automatic test engine 125 to finish the calibration of the RF circuit. In general, in verifying these calibration codes, the code verifier 130 may determine whether the RF circuit configured based on the N calibration codes meets a target set of performance parameters (e.g., a second pass through a measurement system). If the target set of performance parameters is met, the code verifier 130 may write the N calibration codes to the FPROM 140 associated with the RF circuit (also referred to as burning the calibration codes into the FPROM). The calibration codes written to the FPROM 140 may then be used by the RF circuit during operation.

[0019]

[0025] As described, generating and verifying calibration codes for RF circuits can be a resource-intensive process. First, the number of calibration codes generated and verified can increase as the complexity of the RF circuit increases. In addition, for each parameter, calibration can be a multi-step process of generating a calibration code, verifying the calibration code, and refining the generated calibration code based on the output of the verification process. However, it can also be observed that some RF circuit parameters can vary systematically (e.g., can be specific to a die and manufacturing parameters of that die), while other circuit parameters can vary randomly. Because systematically varying parameters can be predictable, aspects of the present disclosure can leverage this predictability to reduce the number of RF circuit calibration parameters generated and verified while maintaining the performance of the RF circuit and minimizing or at least reducing yield loss from circuits that cannot be calibrated to comply with a defined set of performance parameters.

[0020] An Exemplary Machine Learning Model-Based Radio Frequency (RF) Circuit Calibration Pipeline

[0026] To leverage the predictability of some RF circuit calibration parameters, aspects of the present disclosure use machine learning models to identify relationships between different circuit calibration parameters and generate a set of parameters that can predict calibration with minimal or at least reduced yield loss. By doing so, a first set of RF circuit calibration parameters can be generated via a parameter calibrator, and this first set of RF circuit calibration parameters can be used to predict a second set of RF circuit calibration parameters. Because the first set of RF circuit calibration parameters can be a subset of the population of RF circuit calibration parameters, aspects of the present disclosure can reduce computing resources used in calibrating RF circuits by reducing the number of calibration codes that are generated and verified using the calibration pipeline described above.

[0021]

[0027] 2A and 2B show example pipelines 200A and 200B for off-chip calibration of RF circuits using machine learning models according to aspects of the disclosure, in which a code prediction machine learning model can predict values ​​of a set of RF circuit calibration codes based on the output of an off-chip automatic test engine, and can output the predicted values ​​of the set of RF circuit calibration codes for validation or can directly write the predicted values ​​of the set of RF circuit calibration codes to an FPROM or other memory associated with the RF circuit.

[0022]

[0028] 2A illustrates an example pipeline 200A in which RF circuit calibration codes are predicted by a machine learning model. As illustrated, for a set of N RF circuit parameters, RF parameter calibrator 110 can be configured to generate calibration codes for NM parameters. The M parameters can be parameters that can be predicted by code prediction machine learning model 210, as described in further detail below, while the NM parameters can be parameters that are calibrated by RF parameter calibrator 110 and used by code prediction machine learning model 210 to predict calibration codes for the M parameters. The NM calibration codes can be written to a temporary memory 120 on a chip having an RF circuit being calibrated.

[0023]

[0029] Off-chip, the automatic test engine 125 can access these NM calibration codes for evaluation. Additionally, the automatic test engine 125 can provide the NM calibration codes to a code prediction machine learning model 210, which can use at least a portion of the NM calibration codes to predict values ​​of the M calibration codes. The code prediction machine learning model 210 can write the M calibration codes to the temporary memory 120, as shown.

[0024]

[0030] After the code prediction machine learning model 210 generates the M calibration codes and writes the M calibration codes to the temporary memory 120, the code verifier 130 can retrieve the N calibration codes and verify that the codes result in an RF circuit operating according to a defined set of performance parameters. The N calibration codes retrieved from the temporary memory 120 generally include a first subset of RF circuit calibration codes corresponding to the NM parameters calibrated by the RF parameter calibrator 110 and a second subset of RF circuit calibration codes corresponding to the M parameters predicted by the code prediction machine learning model 210. After the code verifier 130 verifies that the N calibration codes result in performance that meets the defined set of performance parameters, the code verifier 130 can write the N calibration codes to the FPROM 140.

[0025]

[0031] In the embodiment shown in pipeline 200A, the number of calibrations performed on the RF circuit parameters may be reduced by M. However, code verifier 130 may verify all N calibration codes (or at least a portion of the N calibration codes, including some of the M calibration codes) before writing the N calibration codes to FPROM 140.

[0026]

[0032] In some aspects, additional performance gains can be achieved by reducing the number of calibration codes verified by the code verifier (and thus undergo refinement via additional calibration). As shown in FIG. 2B, the pipeline 200B allows the M calibration codes predicted by the code prediction machine learning model 210 to be directly written to the FPROM 140 without being verified by the code verifier 130. In the pipeline 200B, the automatic test engine 125 can receive the N M calibration codes generated by the RF parameter calibrator 110 and instruct the code verifier 130 to verify these N M calibration codes. Meanwhile, the code prediction machine learning model 210 can use the N M calibration codes to predict values ​​of calibration codes for the M circuit parameters that were not calibrated by the RF parameter calibrator 110.

[0027]

[0033] Thus, in pipeline 200B, there may be two write paths for writing calibration codes to FPROM 140. A first write path from code verifier 130 may be used to write calibration codes to FPROM 140 for parameters that cannot be predicted (or are designated as not predicted) using code prediction machine learning model 210, and therefore are calibrated using RF parameter calibrator 110, and are verified using code verifier 130. A second write path from code prediction machine learning model 210 may be used to write predicted calibration codes for those parameters that can be predicted (and are designated as predicted) based on the N M calibration codes generated by RF parameter calibrator 110, to FPROM 140. Because M calibration codes may be written directly to FPROM 140 in pipeline 200A without processing these codes through code verifier 130, both the number of calibrations and the number of verifications may be reduced by M, thus further reducing the amount of computing resources used to calibrate the RF circuitry.

[0028]

[0034] In some aspects, RF circuit parameter calibration can be performed on-chip without the use of an off-chip automatic test engine. Figures 3A and 3B show an example pipeline for on-chip calibration of RF circuits using machine learning models according to aspects of the present disclosure.

[0029]

[0035] As shown by pipeline 300A of FIG. 3A, RF parameter calibrator 110 may receive N RF circuit parameters and generate N M calibration codes for a subset of calibration parameters that cannot be predicted (or are designated not to be predicted) using code prediction machine learning model 310. The N M calibration codes may be written to temporary memory 120 and provided to code prediction machine learning model 310. Code prediction machine learning model 310 may use the N M calibration codes as input to predict values ​​of M parameters for which RF parameter calibrator 110 did not generate calibration codes. These M calibration codes may also be written to temporary memory 120, resulting in N calibration codes being stored in temporary memory 120. Thus, similar to pipeline 200A shown in FIG. 2A, code verifier 130 in pipeline 300A may verify the N calibration codes and, once verified, write the N calibration codes to FPROM 140. Thus, similar to pipeline 200A, the number of calibration operations performed in pipeline 300A may be reduced by M, while the number of verification operations may remain at N.

[0030]

[0036] Similar to pipeline 200B, pipeline 300B of FIG. 3B can achieve further reduction in computational resource utilization in calibration and verification of RF circuits. As shown in FIG. 3B, code prediction machine learning model 310 can use NM calibration codes to predict calibration codes for M parameters that were not calibrated by RF parameter calibrator 110. The M calibration codes may be directly written to FPROM 140 (e.g., without being verified by code verifier 130). Meanwhile, the NM calibration codes can be verified by code verifier 130 and, once verified, can be written to FPROM 140. Thus, in pipeline 300B, by directly writing M calibration codes to FPROM 140, the number of calibration operations and the number of verification operations can be reduced by M.

[0031]

[0037] In some aspects, the machine learning model can be used to predict calibration codes for a subset of RF circuit calibration parameters for direct use by the RF circuit. As shown by the pipeline 400 in FIG. 4, NM calibration codes can be generated by the RF parameter calibrator 110 for a subset of parameters that cannot be predicted (or are specified not to be predicted) by the code prediction machine learning model 410. The NM calibration codes can be written to the temporary memory 120 and verified (and possibly refined based on the output of the code verifier 130) by the code verifier 130. The verified NM calibration codes can be written to the FPROM 140.

[0032]

[0038] For the M parameters that can be predicted (and are specified to be predicted) based on the NM calibration codes, the code prediction machine learning model 410 can predict these codes while the RF circuit is operational. Thus, the M calibration codes may be output for use in calibrating the RF circuit along with the NM calibration codes written to the FPROM 140, and the RF circuit can be configured with a total of N codes. Because the M calibration codes can be generated while the RF circuit is operational, the pipeline 400 can reduce the number of calibration operations and the number of verification operations by M. Furthermore, because the NM calibration codes are written to the FPROM 140, the size of the FPROM 140 can also be reduced by a factor of M.

[0033] Exemplary Machine Learning Model for Predicting RF Circuit Calibration Parameters

[0039] The machine learning models described herein can generally be trained to identify a subset of calibration parameters that can predict yield loss below a threshold amount. That is, the calibration parameters predicted by the machine learning models described herein can result in calibration parameters that enable an RF circuit to operate in accordance with a defined set of performance parameters without losing more than a threshold amount of the circuit due to failure to operate in accordance with the defined set of performance parameters. In general, the machine learning models can use as input a set of representative parameters that correspond to parameters calibrated by the RF parameter calibrator 110 shown in FIGS. 2-4 to generate a set of non-representative parameters.

[0034]

[0040] In one embodiment, the machine learning model can be trained based on pairwise correlations between parameters in a dataset of calibration parameters across a set of dies on which RF circuits are calibrated. The pairwise correlations can be clustered into multiple clusters by correlation. In each cluster, one parameter in the representative set of parameters may be associated with multiple parameters in the non-representative set of parameters. The clustering can be optimized to produce a minimum number of clusters. This embodiment can allow some parameters to be predicted, but a minimum number of parameters can be calibrated using values ​​of other parameters while maintaining yield loss below a threshold amount.

[0035]

[0041] In another example, the machine learning model can be trained based on yield similarity clustering. In generating the training data set, a yield loss percentage can be calculated for each pair of parameters in the historical data set of RF circuit calibration parameters. In general, the yield loss can be calculated based on the yield loss experienced by calibrating the second parameter in the pair of parameters using the value of the first parameter. In general, the yield loss similarity can indicate whether the yield loss is maintained below a threshold amount by replacing the value of the second parameter with the value of the first parameter. The yield loss similarity can be clustered such that the centroid of each cluster can correspond to a representative parameter calibrated by the RF parameter calibrator 110 shown in Figures 2-4, while other parameters in the cluster can correspond to non-representative parameters that can be predicted by the machine learning model. The number of clusters can be minimized such that the yield loss of each cluster is below a threshold amount.

[0036]

[0042] To allow further reduction in the number of RF circuit calibration parameters that can be predicted using the machine learning model, an iterative process can be used to evaluate the RF circuit calibration parameters and identify parameters that have a yield loss below a threshold. Thus, a set of sample parameters may be selected using an iterative or "greedy" method, and predicted calibration codes may be generated based on one or more regression models. For example, the regression model may be a lasso regression model in a loss function that is optimized by removing or minimizing features that are less important to the predictions generated by the model. In this case, the circuit calibration may converge to a local optimum rather than a global optimum, and some circuit calibration parameters that can be predicted may not actually be predicted using these techniques.

[0037]

[0043] In yet another embodiment, the machine learning model used to identify and predict values ​​of the RF circuit calibration parameters may be a dropout gradient descent network. A dropout gradient descent network generally allows the machine learning model to use information from each of the sample parameters to predict values ​​of the non-sample parameters, and the set of samples can be minimized based on a sparsity criterion. To build this dropout gradient descent network, which may be a neural network, a sample search space can be first generated for each parameter. For each parameter, the discriminant connections to the parameters that are masked in the neural network may be used to predict values ​​of other parameters in the population of RF circuit calibration parameters. The candidate set of sample parameters may be selected based on, for example, a linear regression model, and the set of candidate non-sample parameters may be a set of parameters that can be predicted with a yield loss below a threshold.

[0038]

[0044] The candidate set of non-sample parameters can be further refined based on a drop probability metric associated with each parameter in the candidate set of non-sample parameters. To do so, the drop probabilities may be initialized based on weights extracted from the linear regression model described above. For each candidate non-sample parameter p ne is the target parameter T e The weight w associated with pneTe may be associated with p ne ≠T e For each parameter, the formula: max(p e ), e∈T, the maximum weight is selected over all target parameters T. p ne The dropout probability of p can be initialized as the output of a softmax function over all parameters p.

[0039]

[0045] The final set of sample parameters can be selected by identifying parameters in the candidate set of non-sample parameters that are actual sample parameters to be calibrated to predict values ​​of the non-sample parameters. To do so, a variational mask can be applied to the input layer to optimize a parameterized dropout rate. A loss function can penalize the count of features that are not removed from the set of candidate non-sample values. Parameters with a keep probability above a threshold can be kept in the set of non-sample parameters, and parameters with a keep probability below the threshold can be included in the set of sample parameters. Finally, based on the learned set of non-sample parameters, a linear network can be trained to identify a learned set of sample parameters. This can result in a model that has no dropout on the non-sample parameters whose values ​​the model is used to predict.

[0040]

[0046] In another example, the dropout gradient descent network can be structured as a linear network, such that a candidate set of predictable parameters can be generated based on gradient descent optimization of a linear regression function, which can be a regression function for weights associated with each parameter in the candidate set.

[0041] Exemplary RF Circuit Calibration Using Machine Learning Models

[0047] FIG. 5 illustrates example operations 500 that may be performed (e.g., by RF circuit calibration pipelines 200A-B, 300A-B, or 400 shown in FIGS. 2-4) to calibrate an RF circuit using a machine learning model.

[0042]

[0048] As shown, the operations 500 may begin at block 510 where a first set of RF circuit calibration parameters are calibrated. The first subset of RF circuit calibration parameters may be a set of representative parameters that can be used by a machine learning model to predict values ​​for one or more parameters in a second set of RF circuit calibration parameters (also referred to as a set of non-represented parameters).

[0043]

[0049] At block 520, values ​​of a second subset of the RF circuit calibration parameters are predicted based on the machine learning model and the first subset of the RF circuit calibration parameters. The second subset of the RF circuit calibration parameters may include, for example, RF circuit calibration parameters that are predictive with yield loss below a threshold.

[0044]

[0050] In some aspects, the second subset of RF circuit calibration parameters can be identified using various techniques. For example, to identify the second subset of RF circuit calibration parameters, pairwise correlations between parameters in the historical data set of RF circuit calibration parameters can be extracted. These pairwise correlations can be clustered into a plurality of clusters. Each cluster of the plurality of clusters may generally be associated with a plurality of parameters in the second subset of RF circuit calibration parameters and a parameter in the first subset of RF circuit calibration parameters. The clustering may be performed such that a minimum number of clusters is generated.

[0045]

[0051] In some aspects, identifying the second subset of RF circuit calibration parameters may be performed for each parameter pair. For each pair of parameters including a first parameter and a second parameter in the historical dataset of RF circuit calibration parameters, a yield loss may be calculated by calibrating the second parameter using a value of the first parameter. A yield loss similarity value may be generated for each pair of parameters based on the yield loss calculated for each pair of parameters, and pairwise correlations may be clustered between each respective pair of parameters based on the yield loss similarity value for each respective pair of parameters. The clustering may be performed such that a minimum number of clusters are generated.

[0046]

[0052] In some aspects, identifying the second subset of RF circuit calibration parameters may include iteratively evaluating the RF circuit calibration parameters to identify parameters having a yield loss below a threshold.

[0047]

[0053] In some aspects, identifying the second subset of RF circuit calibration parameters can include identifying the second subset of RF circuit calibration parameters based on a dropout gradient descent network. The dropout gradient descent network can include, for example, a neural network. In such a case, the second subset of RF circuit calibration parameters can be identified on a parameter-by-parameter basis. For each parameter in the population of RF circuit calibration parameters, other parameters in the population of RF circuit calibration parameters are predicted using the discriminatory connections to the respective parameters masked in the neural network. A candidate set of predictable parameters is generated based on predicted parameters having a yield loss below a threshold, and the candidate set of predictable parameters is refined based on a drop probability metric associated with each parameter in the candidate set of predictable parameters.

[0048]

[0054] In some aspects, to generate a candidate set of predictable parameters, a weight can be assigned to each respective parameter in the candidate set based on the weights extracted from the neural network. In general, the weight associated with each respective parameter corresponds to the influence of the respective parameter on the respective target parameter. A maximum weight for each respective parameter in the candidate set is identified over the set of target parameters associated with the respective parameter. A drop probability value is initialized for each respective parameter based on a softmax function calculated for the maximum weight for each respective parameter in the candidate set.

[0049]

[0055] In some aspects, to generate a candidate set of predictable parameters, the candidate set is generated based on a gradient descent optimization of a linear regression function over a weight associated with each respective parameter in the candidate set.

[0050]

[0056] At block 530, at least a first subset of the RF circuit calibration parameters is verified.

[0051]

[0057] In block 540, after verifying at least the first subset of the RF circuit calibration parameters, the at least the first subset of the RF circuit calibration parameters is written to a memory associated with the RF circuit (e.g., FPROM 140).

[0052]

[0058] In some aspects, the second subset of the RF circuit calibration parameters may also be verified. After verifying the second subset of the RF circuit calibration parameters, the second subset of the RF circuit calibration parameters may be written to a memory associated with the RF circuit.

[0053]

[0059] In some aspects, the RF circuit can operate based on at least the first subset of the RF circuit calibration parameters and a second subset of the RF circuit calibration parameters written to a memory associated with the RF circuit, The second subset of the RF circuit parameters may be used to operate the RF circuit directly without going through a verification stage or may be verified before being used to operate the RF circuit.

[0054] Exemplary Processing System for Calibrating RF Circuits Using Machine Learning Models

[0060] FIG. 6 illustrates an example processing system 600 for calibrating RF circuits using machine learning models, such as, for example, as described herein with respect to FIG.

[0055]

[0061] The processing system 600 includes a central processing unit (CPU) 602, which in some examples may be a multi-core CPU. Instructions executed in the CPU 602 may be loaded from a program memory associated with the CPU 602 or may be loaded from a memory 624, for example.

[0056]

[0062] The processing system 600 also includes additional processing components tailored to specific functions, such as a graphics processing unit (GPU) 604, a digital signal processor (DSP) 606, a neural processing unit (NPU) 608, a multimedia processing unit 610, and wireless connectivity components 612.

[0057]

[0063] An NPU, such as NPU 608, is generally a specialized circuit configured to implement all the necessary control and computational logic to execute machine learning algorithms, such as algorithms for processing artificial neural networks (ANNs), deep neural networks (DNNs), random forests (RFs), etc. An NPU may alternatively be referred to as a neural signal processor (NSP), a tensor processing unit (TPU), a neural network processor (NNP), an intelligence processing unit (IPU), a vision processing unit (VPU), or a graph processing unit.

[0058]

[0064] An NPU, such as NPU 608, is configured to accelerate the execution of common machine learning tasks, such as image classification, machine translation, object detection, and various other predictive models. In some embodiments, multiple NPUs may be instantiated on a single chip, such as a system on a chip (SoC), while in other embodiments, an NPU may be part of a dedicated neural network accelerator.

[0059]

[0065] NPUs can be optimized for training or inference, or in some cases can be configured to balance performance between both. For NPUs capable of performing both training and inference, the two tasks can still generally be performed independently.

[0060]

[0066] NPUs designed to accelerate training are generally configured to accelerate the optimization of new models, which is a highly computationally intensive operation that involves inputting an existing dataset (often labeled or tagged), iterating over the dataset, and then adjusting model parameters, such as weights and biases, to improve model performance. In general, optimization based on mispredictions involves backpropagating through layers of the model to determine gradients to reduce prediction errors.

[0061]

[0067] NPUs designed to accelerate inference are generally configured to operate on complete models, such that such NPUs can be configured to input new data and rapidly process that data through models that have already been trained to produce model outputs (e.g., inferences).

[0062]

[0068] In one implementation, the NPU 608 is part of one or more of the CPU 602, the GPU 604, and / or the DSP 606.

[0063]

[0069] The processing system 600 may also include one or more input and / or output devices 622, such as a screen, a touch-sensitive surface (including a touch-sensitive display), physical buttons, a speaker, a microphone, etc.

[0064]

[0070] In some embodiments, one or more of the processors of processing system 600 may be based on the ARM or RISC-V instruction set.

[0065]

[0071] Processing system 600 also includes memory 624, which represents one or more static and / or dynamic memories, such as dynamic random access memory, flash-based static memory, etc. In this example, memory 624 includes computer-executable components capable of being executed by one or more of the aforementioned processors of processing system 600.

[0066]

[0072] In particular, in this embodiment, memory 624 includes an RF circuit calibration component 624A, a parameter value prediction component 624B, a parameter verification component 624C, and a parameter writing component 624D. The components shown, and other components not shown, may be configured to perform various aspects of the methods described herein.

[0067] Example clauses

[0073] Clause 1: A method for calibrating a radio frequency (RF) circuit, comprising: calibrating a first subset of RF circuit calibration parameters; predicting values ​​of a second subset of the RF circuit calibration parameters, different from the first subset of the RF circuit calibration parameters, based on a machine learning model and the first subset of the RF circuit calibration parameters; validating at least the first subset of the RF circuit calibration parameters; and after validation, writing at least the first subset of the RF circuit calibration parameters to a memory associated with the RF circuit.

[0068]

[0074] Clause 2: The method of clause 1, further comprising identifying a second subset of RF circuit calibration parameters.

[0069]

[0075] Clause 3: The method of clause 2, wherein identifying the second subset of RF circuit calibration parameters includes extracting pair-wise correlations between parameters in a historical dataset of RF circuit calibration parameters, and clustering the pair-wise correlations into a plurality of clusters, each cluster of the plurality of clusters being associated with a plurality of parameters in the second subset of RF circuit calibration parameters and a parameter in the first subset of RF circuit calibration parameters.

[0070]

[0076] Clause 4: The method of clause 3, wherein clustering the pairwise correlations includes generating a minimum number of clusters.

[0071]

[0077] Clause 5: The method of any one of clauses 2 to 4, wherein identifying a second subset of RF circuit calibration parameters includes: for each respective pair of parameters including a first parameter and a second parameter in the historical dataset of RF circuit calibration parameters, calculating a yield loss produced by calibrating the second parameter using a value of the first parameter; generating a yield loss similarity value for each respective pair of parameters based on the yield loss calculated for each respective pair of parameters; and clustering pair-wise correlations between each respective pair of parameters based on the yield loss similarity value for each respective pair of parameters.

[0072]

[0078] Clause 6: The method of clause 5, wherein clustering the pair-wise correlations includes generating a minimum number of clusters such that a yield loss for each cluster is less than a threshold.

[0073]

[0079] Clause 7: The method of any one of clauses 2 to 6, wherein identifying a second subset of RF circuit calibration parameters includes iteratively evaluating the RF circuit calibration parameters to identify parameters having a yield loss below a threshold.

[0074]

[0080] Clause 8: The method of any one of clauses 2 to 7, wherein identifying the second subset of RF circuit calibration parameters includes identifying the second subset of RF circuit calibration parameters based on a dropout gradient descent network.

[0075]

[0081] Clause 9: The method of clause 8, wherein the dropout gradient descent network includes a neural network, and identifying the second subset of RF circuit calibration parameters includes: for each respective parameter in the population of RF circuit calibration parameters, predicting other parameters in the population of RF circuit calibration parameters using discriminatory connections to the respective parameters masked in the neural network; generating a candidate set of predictable parameters based on the predicted parameters having yield loss below a threshold; and refining the candidate set of predictable parameters based on a drop probability metric associated with each parameter in the candidate set of predictable parameters.

[0076]

[0082] Clause 10: The method of clause 9, wherein generating a candidate set of predictable parameters includes assigning a weight to each respective parameter in the candidate set based on weights extracted from the neural network, the weight of each respective parameter corresponding to an influence of the respective parameter on a respective target parameter; identifying a maximum weight for each respective parameter in the candidate set across the set of target parameters associated with the respective parameters; and initializing a drop probability value for each respective parameter based on a softmax function calculated for the maximum weight for each respective parameter in the candidate set.

[0077]

[0083] Clause 11: The method of clause 9 or 10, wherein generating a candidate set of predictable parameters includes generating the candidate set based on gradient descent optimization of a linear regression function for a weight associated with each respective parameter in the candidate set.

[0078]

[0084] Clause 12. The method of any one of clauses 1 to 11, wherein the second subset of RF circuit calibration parameters includes RF circuit calibration parameters that are predictive of yield loss below a threshold.

[0079]

[0085] Clause 13: The method of any one of clauses 1 to 12, further comprising verifying a second subset of RF circuit calibration parameters.

[0080]

[0086] Clause 14: The method of clause 13, further comprising, after verifying the second subset of RF circuit calibration parameters, writing the second subset of RF circuit calibration parameters to a memory associated with the RF circuit.

[0081]

[0087] Clause 15: A method according to any one of clauses 1 to 14, further comprising operating the RF circuit based on at least the first subset of RF circuit calibration parameters and a second subset of RF circuit calibration parameters written to a memory associated with the RF circuit.

[0082]

[0088] Clause 16: An apparatus comprising: a memory having executable instructions stored therein; and a processor configured to execute the executable instructions to cause the apparatus to perform a method according to any one of clauses 1 to 15.

[0083]

[0089] Clause 16: Apparatus comprising means for carrying out the method according to any one of clauses 1 to 15.

[0084]

[0090] Clause 17: A non-transitory computer-readable medium having stored thereon instructions that, when executed by a processor, cause the processor to perform the method of any one of clauses 1 to 15.

[0085]

[0091] Clause 18: A computer program product embodied on a computer-readable storage medium comprising code for performing the method of any one of clauses 1 to 15.

[0086] Additional Considerations

[0092] The foregoing description is provided to enable any person skilled in the art to practice the various embodiments described herein. The examples described herein are not intended to limit the scope, applicability, or embodiments described in the claims. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments. For example, changes may be made in the function and arrangement of the elements described without departing from the scope of the disclosure. Various embodiments may omit, substitute, or add various procedures or components as appropriate. For example, the methods described may be performed in an order different from that described, and various steps may be added, omitted, or combined. Also, features described with respect to some embodiments may be combined in some other embodiments. For example, an apparatus may be implemented or a method may be practiced using any number of aspects described herein. Furthermore, the scope of the disclosure is intended to encompass apparatuses or methods that are implemented using other structures, functions, or structures and functions in addition to or other than the various aspects of the disclosure described herein. It should be understood that any aspect of the disclosure disclosed herein may be embodied by one or more elements of a claim.

[0087]

[0093] As used herein, the word "exemplary" means "serving as an example, instance, or illustration." Any aspect described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other aspects.

[0088]

[0094] As used herein, a phrase referring to "at least one of" a list of items refers to any combination of those items, including single members. By way of example, "at least one of a, b, or c" is intended to encompass a, b, c, ab, ac, bc, and abc, as well as any combination having multiples of the same element (e.g., aa, aaa, aab, aac, abb, acc, bb, bbb, bbc, cc, and ccc, or any other permutation of a, b, and c).

[0089]

[0095] As used herein, the term "determining" encompasses a wide variety of actions. For example, "determining" can include calculating, computing, processing, deriving, investigating, searching (e.g., searching a table, database, or another data structure), ascertaining, and the like. "Determining" can also include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), and the like. "Determining" can also include resolving, selecting, choosing, establishing, and the like.

[0090]

[0096] The methods disclosed herein include one or more steps or actions for achieving the method. The steps and / or actions of those methods may be interchanged with one another without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order of specific steps and / or actions and / or the use of those steps and / or actions may be modified without departing from the scope of the claims. Furthermore, various operations of the methods described above may be performed by any suitable means capable of performing the corresponding functions. Those means may include various hardware and / or software component(s) including, but not limited to, circuits, application specific integrated circuits (ASICs), or processors, and / or various hardware and / or software module(s). In general, when operations are illustrated in figures, those operations may have corresponding equivalent means-plus-function components that are similarly numbered.

[0091]

[0097] An exemplary means for calibrating may include, for example, the RF parameter calibrator 110 shown in Figures 2-4, among others. An exemplary means for predicting may include, for example, the code prediction machine learning model 210, 310, or 410 shown in Figures 2-4, among others. An exemplary means for verifying may include, for example, the code verifier 130 shown in Figures 2-4, among others. An exemplary means for writing may include, for example, the code verifier 130, the code prediction machine learning model 210, 310, or 410, and / or the FPROM 140 shown in Figures 2-4, among others.

[0092]

[0098] The following claims are not intended to be limited to the embodiments set forth herein, but are to be accorded the full scope consistent with the language of the claims. Within the claims, reference to an element by the singular is not intended to mean "only one" unless expressly stated as such, but rather "one or more." The term "several" refers to one or more, unless expressly stated otherwise. No element of a claim is to be construed under the provisions of 35 U.S.C. 112(f) unless the element is expressly recited using the phrase "means for" or, in the case of a method claim, unless the element is recited using the phrase "step for." All structural and functional equivalents of the elements of the various aspects described throughout this disclosure that are known or later become known to those skilled in the art are expressly incorporated herein by reference and are intended to be encompassed by the claims. Furthermore, nothing disclosed herein is intended to be made public, regardless of whether such disclosure is expressly recited in the claims.

Claims

1. 1. A method for calibrating a radio frequency (RF) circuit, comprising: calibrating a first subset of RF circuit calibration parameters; predicting values ​​of a second subset of RF circuit calibration parameters, the second subset being different from the first subset of RF circuit calibration parameters, based on the machine learning model and the first subset of RF circuit calibration parameters; verifying at least the first subset of RF circuit calibration parameters; after said verifying, writing at least said first subset of RF circuit calibration parameters to a memory associated with said RF circuit; A method comprising:

2. The method of claim 1 , further comprising identifying the second subset of RF circuit calibration parameters.

3. identifying the second subset of RF circuit calibration parameters; Extracting pairwise correlations between parameters in a historical data set of RF circuit calibration parameters; clustering the pair-wise correlations into a plurality of clusters, each cluster of the plurality of clusters being associated with a plurality of parameters in the second subset of RF circuit calibration parameters and a parameter in the first subset of RF circuit calibration parameters; Including, The method of claim 2.

4. The method of claim 3 , wherein clustering the pair-wise correlations comprises creating a minimum number of clusters.

5. identifying the second subset of RF circuit calibration parameters; For each respective pair of parameters including a first parameter and a second parameter in the historical data set of RF circuit calibration parameters: calculating a yield loss produced by calibrating the second parameter using the value of the first parameter; generating a yield loss similarity value for each pair of parameters based on the calculated yield loss for each pair of parameters; clustering pairwise correlations between each respective pair of parameters based on the yield loss similarity value for each respective pair of parameters; Including, The method of claim 2.

6. The method of claim 5 , wherein clustering the pair-wise correlations comprises creating a minimum number of clusters such that each cluster has a yield loss below a threshold.

7. 3. The method of claim 2, wherein identifying the second subset of RF circuit calibration parameters comprises iteratively evaluating RF circuit calibration parameters to identify parameters having a yield loss below a threshold.

8. 3. The method of claim 2, wherein identifying the second subset of RF circuit calibration parameters comprises identifying the second subset of RF circuit calibration parameters based on a dropout gradient descent network.

9. the dropout gradient descent network comprises a neural network; Identifying the second subset of RF circuit calibration parameters includes, for each respective parameter of the population of RF circuit calibration parameters: predicting other parameters in the population of RF circuit calibration parameters using discriminatory connections for each of the masked parameters in the neural network; generating a candidate set of predictable parameters based on the predicted parameters having a yield loss below a threshold; refining the candidate set of predictable parameters based on a drop probability metric associated with each parameter in the candidate set of predictable parameters; Including, The method according to claim 8.

10. Generating the candidate set of predictable parameters assigning a weight to each respective parameter in the candidate set based on weights extracted from the neural network, the weight of each respective parameter corresponding to an influence of the respective parameter on a respective target parameter; identifying a maximum weight for each respective parameter in the candidate set over a set of target parameters associated with the respective parameter; initializing a drop probability value for each respective parameter based on a softmax function calculated for the maximum weight for each respective parameter in the candidate set; Including, 10. The method of claim 9.

11. 10. The method of claim 9, wherein generating the candidate set of predictable parameters comprises generating the candidate set based on a gradient descent optimization of a linear regression function for a weight associated with each respective parameter in the candidate set.

12. The method of claim 1 , wherein the second subset of RF circuit calibration parameters comprises RF circuit calibration parameters that are predictive of yield loss below a threshold.

13. The method of claim 1 , further comprising verifying the second subset of RF circuit calibration parameters.

14. 14. The method of claim 13, further comprising, after verifying the second subset of RF circuit calibration parameters, writing the second subset of RF circuit calibration parameters to the memory associated with the RF circuit.

15. 2. The method of claim 1, further comprising: operating the RF circuit based on the second subset of RF circuit calibration parameters and the at least the first subset of RF circuit calibration parameters written to the memory associated with the RF circuit.

16. 1. An apparatus for wireless communication, comprising: a radio frequency (RF) circuit having a first memory; a second memory having executable instructions stored therein; a processor configured to execute the executable instructions, Calibrating a first subset of RF circuit calibration parameters; predicting values ​​of a second subset of RF circuit calibration parameters, the second subset being different from the first subset of RF circuit calibration parameters based on the machine learning model and the first subset of RF circuit calibration parameters; verifying at least the first subset of RF circuit calibration parameters; after said verifying, writing at least said first subset of RF circuit calibration parameters to said first memory; a processor for An apparatus comprising:

17. 17. The apparatus of claim 16, wherein the processor is further configured to identify the second subset of RF circuit calibration parameters.

18. To identify the second subset of RF circuit calibration parameters, the processor: Extracting pairwise correlations between parameters in a historical data set of RF circuit calibration parameters; clustering the pair-wise correlations into a plurality of clusters, each cluster of the plurality of clusters being associated with a plurality of parameters in the second subset of RF circuit calibration parameters and a parameter in the first subset of RF circuit calibration parameters. It is configured as follows:

20. The apparatus of claim 17.

19. To identify the second subset of RF circuit calibration parameters, the processor: For each respective pair of parameters including a first parameter and a second parameter in the historical data set of RF circuit calibration parameters: Calculating a yield loss produced by calibrating the second parameter using the value of the first parameter; generating a yield loss similarity value for each pair of parameters based on the calculated yield loss for each pair of parameters; clustering pairwise correlations between each respective pair of parameters based on the yield loss similarity value for each respective pair of parameters. It is configured as follows:

20. The apparatus of claim 17.

20. 20. The apparatus of claim 19, wherein to cluster the pair-wise correlations, the processor is configured to generate a minimum number of clusters such that each cluster has a yield loss below a threshold.

21. 20. The apparatus of claim 17, wherein to identify the second subset of RF circuit calibration parameters, the processor is configured to iteratively evaluate RF circuit calibration parameters to identify parameters having a yield loss below a threshold.

22. 20. The apparatus of claim 17, wherein to identify the second subset of RF circuit calibration parameters, the processor is configured to identify the second subset of RF circuit calibration parameters based on a dropout gradient descent network.

23. the dropout gradient descent network comprises a neural network; To identify the second subset of RF circuit calibration parameters, the processor performs, for each respective parameter of a population of RF circuit calibration parameters: predicting other parameters in the population of RF circuit calibration parameters using discriminatory connections for each of the masked parameters in the neural network; generating a candidate set of predictable parameters based on the predicted parameters having a yield loss below a threshold; refining the candidate set of predictable parameters based on a drop probability metric associated with each parameter in the candidate set of predictable parameters. It is configured as follows:

23. The apparatus of claim 22.

24. To generate the candidate set of predictable parameters, the processor: assigning a weight to each respective parameter in the candidate set based on weights extracted from the neural network, the weight of each respective parameter corresponding to an influence of the respective parameter on a respective target parameter; identifying a maximum weight for each respective parameter in the candidate set across a set of target parameters associated with the respective parameter; initializing a drop probability value for each respective parameter based on a softmax function calculated for the maximum weight for each respective parameter in the candidate set; It is configured as follows:

24. The apparatus of claim 23.

25. 24. The apparatus of claim 23, wherein to generate the candidate set of predictable parameters, the processor is configured to generate the candidate set based on a gradient descent optimization of a linear regression function for a weight associated with each respective parameter in the candidate set.

26. 17. The apparatus of claim 16, wherein the second subset of RF circuit calibration parameters comprises RF circuit calibration parameters that are predictive of yield loss below a threshold.

27. The processor, verifying the second subset of RF circuit calibration parameters; writing the second subset of RF circuit calibration parameters to the first memory after verifying the second subset of RF circuit calibration parameters; The method is further configured as follows:

17. The apparatus of claim 16.

28. 17. The apparatus of claim 16, wherein the processor is further configured to operate the RF circuit based on the second subset of RF circuit calibration parameters and the at least the first subset of RF circuit calibration parameters written to the first memory.

29. 1. An apparatus for calibrating a radio frequency (RF) circuit, comprising: means for calibrating a first subset of RF circuit calibration parameters; means for predicting values ​​of a second subset of RF circuit calibration parameters, the second subset being different from the first subset of RF circuit calibration parameters, based on the machine learning model and the first subset of RF circuit calibration parameters; means for verifying at least the first subset of RF circuit calibration parameters; means for writing at least the first subset of RF circuit calibration parameters to a memory associated with the RF circuit after said verification; An apparatus comprising:

30. 1. A non-transitory computer-readable medium having stored thereon instructions that, when executed by a processor, cause the processor to perform operations for calibrating a radio frequency (RF) circuit, the operations comprising: calibrating a first subset of RF circuit calibration parameters; predicting values ​​of a second subset of RF circuit calibration parameters, the second subset being different from the first subset of RF circuit calibration parameters, based on the machine learning model and the first subset of RF circuit calibration parameters; verifying at least the first subset of RF circuit calibration parameters; after said verifying, writing at least said first subset of RF circuit calibration parameters to a memory associated with said RF circuit; Including, Non-transitory computer-readable medium.

Images