Method for digital predistortion, linearization system for performing digital predistortion, and non-transitory machine-readable medium

By configuring a digital predistortion system and utilizing diverse metrics to select data samples and optimize the basis function set, the adaptive compensation problem under rapidly changing operating conditions in wireless communication is solved, achieving high-performance digital predistortion under rapidly changing conditions and improving communication quality.

CN112640299BActive Publication Date: 2024-11-12NANOSEMI INC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201980049576.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2018-05-25
Filing Date
2019-05-24
Publication Date
2024-11-12
Estimated Expiration
2039-05-24

AI Technical Summary

Technical Problem

Existing digital predistortion techniques struggle to achieve high-performance adaptive compensation under rapidly changing operating conditions, especially in wireless communication, particularly in bidirectional communication between user equipment (UE) and base stations, where they face the problem of nonlinear characteristics caused by rapidly changing frequency, power, and environmental parameters.

Method used

By configuring a digital predistortion system, selecting data samples using diverse metrics, and optimizing the basis function set, a fast and adaptive digital predistortion method can be achieved to adapt to changes in the operating conditions of wireless devices, including rapid changes in parameters such as temperature, average power, and VSWR.

Benefits of technology

This technology improves the robustness and accuracy of digital predistortion under rapidly changing operating conditions, reduces the impact of nonlinear characteristics, and enhances communication quality.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112640299B_ABST
    Figure CN112640299B_ABST
Patent Text Reader

Abstract

A digital predistortion implementation is disclosed, including a method, the method comprising collecting data for a wireless device having a transmission chain, the transmission chain comprising at least one power amplifier that produces an output having nonlinear distortion, the data comprising data records, wherein each data record comprises input signal data and corresponding output signal data for the transmission chain, and wherein each record is also associated with a set of operating condition values, each data record is associated with a diversity measure representing a degree of distinguishability of the record from other data records. The method also includes selecting at least some of the data records based on a diversity criterion calculated based on the diversity measure for each of the records; calculating digital predistortion parameters for controlling a digital predistorter of the wireless device based on the selected records; and applying the digital predistorter configured with the calculated digital predistortion parameters to a subsequent input signal.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of U.S. Provisional Application No. 62 / 676,617, entitled “Parameterized Universal DPD Compensators,” U.S. Provisional Application No. 62 / 676,677, entitled “Digital Predistortion System with Adaptability to Fast Abrupt Changes of Environmental Characteristics,” and U.S. Provisional Application No. 62 / 676,682, entitled “Digital Predistortion System Implementation Using Data Samples Selected According to Variety Metrics,” all filed on May 25, 2018, the entire contents of which are incorporated herein by reference in their entirety. Background Art

[0003] The present disclosure relates to methods of digital predistortion, and more particularly to one or more techniques that may be applied alone or in combination to digital predistortion under rapidly changing operating conditions.

[0004] Power amplifiers, particularly those used to transmit radio frequency communications, generally have nonlinear characteristics. For example, when the output power of a power amplifier approaches its maximum rated output, nonlinear distortion of the output occurs. One method of compensating for the nonlinear characteristics of a power amplifier is to "pre-distort" the input signal to eliminate the nonlinearity of the power amplifier before providing it to the power amplifier (e.g., by adding "inverse distortion" to the input signal). The final output of the power amplifier is a linear amplification of the input signal with reduced nonlinear distortion. Digital predistortion devices (including, for example, power amplifiers) are relatively cheap and power efficient. These attributes make digital predistortion power amplifiers attractive for use in telecommunication systems, where the amplifier is required to reproduce the signal presented at its input cheaply, efficiently, and accurately. However, the computational workload associated with conventional digital predistortion (DPD) adaptive processes is substantial, which adversely affects predistortion accuracy and robustness because the computational workload may impose limits on the number and speed at which DPD coefficients can be derived.

[0005] There are several challenges for DPD adaptive implementation involving bidirectional wireless communications (e.g., between a base station (BS) and a user equipment (UE)). For example, a UE typically transmits (uploads data) on demand and only with sufficient output power to save battery power. For example, the adjustment of the output power level is achieved by changing the baseband input level, RF gain, and supply voltage of the power amplifier. In addition, the UE transmitter can be controlled (e.g., under the command of the BS) to transmit using a changing frequency range (e.g., channel). Further, operating modes and parameters (such as power level, supply voltage level, output load of the PA (e.g., VSWR)) can be changed rapidly in the UE device. The combination of such changes in operating conditions can change the characteristics of the transmit chain, and more specifically change the nonlinear properties of the transmit chain that is being compensated by the DPD.

[0006] Therefore, there is a need for a DPD method that can provide high performance DPD (eg, as measured by linearity, error rate, spectral mask violations, etc.) over a wide range of operating conditions and that can adapt to such rapid changes in operating conditions. Summary of the invention

[0007] An adaptive digital predistortion implementation is disclosed that allows a wireless device (typically a UE, but also a base station) to perform fast adaptation (μs level) of its digital predistortion function in response to changes in its operating conditions / points (including changes in environmental conditions). Changes in the operating conditions of the wireless device may occur due to, for example: 1) subcarriers within a component carrier change frequently (e.g., every time slot, where such a time slot has a duration of tens of μs) and their spectral content may jump rapidly in the spectrum; 2) the UE transmitter needs to adjust the average output power by increasing or decreasing the DC supply voltage; 3) changes in other environmental parameters such as temperature and VSWR (such changes may also occur frequently); and / or 4) changes due to the fact that the 5G mmWave spectrum band is much wider than any existing sub-6 GHz band and its spectrum ownership is split among many operators (operators operate different bands within this spectrum range and may use different technologies). Regarding the latter factor, it is noted that for the new mmWave spectrum bands, the frequency coverage of UE devices easily exceeds 1 GHz, and as the UE moves from one cell to another, or even when it is allocated different cellular resources within a particular cell (by a particular operator), the UE operating point may change frequently and abruptly, thus requiring its digital pre-distortion functionality to change according to the changes in the operating point.

[0008] To address the various issues discussed above, several schemes and methods are proposed that facilitate fast, reliable, and robust digital predistortion adaptation techniques. In one example method, a digital predistortion implementation configured for adaptability to fast / sudden changes in spectral localization, average power, VDD, temperature, VSWR, etc. is provided. Such adaptability to fast / sudden changes in the operating point of the wireless device can be achieved, for example, by a DPD system implementation that is configured to derive an adaptive extended set of basis functions from a first set of DPD basis functions based on operating conditions associated with the wireless device and / or the predistorter system.

[0009] In another example method, a wireless communication device having a universal digital predistortion system is implemented, the wireless communication device being configured to collect training samples from multiple narrow RF bands within a wideband RF spectrum range and optimize a universal digital predistorter operable within the wideband RF spectrum range.

[0010] In another example method that facilitates fast and robust adaptation of a digital predistortion function, a digital predistortion (DPD) system uses data samples selected according to a diversity metric to implement digital predistortion adaptation. The system can be configured to collect data distributed over a channel (e.g., for storage in a diversity matrix), select a subset of the data that is more evenly distributed than at least some other subsets of the data, and calculate digital predistortion parameters based on the selected subset of data to weight basis functions used to control digital predistortion operations.

[0011] Thus, in some variations, a method for digital predistortion is provided, the method comprising collecting data for a wireless device comprising a transmission chain comprising at least one power amplifier that produces an output having nonlinear distortion. The data comprises data records, wherein each data record comprises a data record of input signal data and corresponding output signal data for the transmission chain and is also associated with a set of operating condition values. Each data record is also associated with a diversity measure representing a degree of distinguishability of the corresponding data record from other collected data records. The method also comprises selecting at least some of the data records according to a diversity criterion calculated based on the diversity measure for each data record, calculating digital predistortion parameters for controlling a digital predistorter of the wireless device based on the selected at least some of the data records, and applying the digital predistorter configured with the calculated digital predistortion parameters to a subsequent input signal.

[0012] Embodiments of the method may include at least some of the features described in this disclosure, including one or more of the following features.

[0013] The set of operating conditions for each data record may include one or more of the following, for example, channel information associated with a corresponding channel used by the wireless device to process the corresponding data record, a corresponding temperature, a corresponding average power, a corresponding supply voltage (V ) associated with at least one power amplifier of the transmit chain, and a corresponding power supply voltage (V ) associated with at least one power amplifier of the transmit chain. DD ), a corresponding voltage standing wave ratio (VSWR) associated with the wireless device, and / or the age of each data record.

[0014] Collecting data for the wireless device may include collecting data in one or more spectrum channels, wherein the one or more spectrum channels are located within a wideband spectrum range, and wherein each of the one or more spectrum channels has a corresponding bandwidth that is less than a range bandwidth of the spectrum range.

[0015] Each of the data records may include one or more samples obtained by the wireless device under a particular set of operating conditions.

[0016] Each of the data records may be maintained in a separate column of a record matrix data structure.

[0017] Selecting at least some of the data records according to a diversity criterion calculated based on a diversity measure for each data record may include selecting data records associated with respective diversity measures indicating high distinguishability to increase diversity of data records used for calculating digital predistortion parameters.

[0018] The method may further include calculating, for each data record, the diversity measure based on a distance metric, wherein the distance metric represents a distance between a data element of a particular data record and corresponding data elements of at least some of the collected data records.

[0019] The method may also include maintaining the calculated diversity measure for each data record in a distance vector data structure separate from a data structure used to store the collected data records.

[0020] The method may also include collecting additional data records; in response to determining that a diversity measure calculated for a particular additional data record in the collected additional data records is more distinguishable than data records in the previously stored data records relative to previously stored data records available at the record data structure, storing the particular additional data record in the collected additional data records in the record data structure; and recalculating digital predistortion parameters at regular or irregular intervals based on stored data samples available in the record data structure, the stored data samples including the particular additional data record stored in the collected additional data records.

[0021] Storing a particular one of the collected additional data records may include replacing a data record in a previously stored data record with the particular one of the collected additional data records in the record data structure.

[0022] The method may also include calculating a corresponding diversity measure for a particular additional data record among the collected additional data records, including determining a maximum absolute value of differences between corresponding elements of the data records for each pair of records including the particular additional data record among the collected additional data records and at least one data record among the previously stored data records.

[0023] Applying a digital predistorter configured with digital predistortion parameters may include weighting basis functions of the digital predistorter, wherein the basis functions include complex terms applied to values ​​of input samples provided in a data record, and wherein at least some of the basis functions also include real terms applied to values ​​of a set of operating conditions associated with the data record.

[0024] The method may further include deriving basis functions of a digital predistorter from the first basis function set, the basis functions of the digital predistorter being a basis function extension set including (d+1) groups of basis functions, wherein d represents the number of parameters of the operating conditions, wherein a first group of the basis function extension set includes the first basis function set without any modification, and wherein each of the remaining d groups of the basis function extension set is determined as a product of the first basis function set and one of the parameters of the operating conditions.

[0025] In some variations, a linearization system for performing digital predistortion is provided, the linearization system comprising: a transmission chain including at least one power amplifier that produces an output having nonlinear distortion; and a digital predistorter including an adaptive section and a compensator. The digital predistorter is configured to collect data including data records, wherein each data record includes input signal data and corresponding output signal data for the transmission chain and is also associated with a set of operating condition values, wherein each data record is associated with a diversity measure representing a degree of distinguishability of the corresponding data record from other collected data records. The digital predistorter is also configured to select at least some of the data records based on a diversity criterion calculated based on the diversity measure of each data record, calculate digital predistortion parameters for controlling the digital predistorter based on the selected at least some data records, and digitally predistort a subsequent input signal based on the calculated digital predistortion parameters.

[0026] Embodiments of the linearization system may include at least some of the features described in this disclosure, including any of the above method features and one or more of the following features.

[0027] A digital predistorter configured to select at least some of the data records according to a diversity criterion calculated based on a diversity measure for each data record may be configured to select data records associated with respective diversity measures indicating high distinguishability in order to increase the diversity of data records used for calculating digital predistortion parameters.

[0028] The digital predistorter may be further configured to calculate the diversity measure for each data record based on a distance metric, wherein the distance metric represents a distance between a data element of a particular data record and corresponding data elements of at least some of the collected data records.

[0029] The digital predistorter may also be configured to: collect additional data records; in response to determining that a diversity measure calculated for a particular additional data record in the collected additional data records is more distinguishable than a data record in the previously stored data records relative to previously stored data records available at the record data structure, store a particular additional data record in the collected additional data records in the record data structure; and recalculate digital predistortion parameters at regular or irregular intervals based on stored data samples available in the record data structure, the stored data samples including the particular additional data record stored in the collected additional data records.

[0030] The digital predistorter configured to perform digital predistortion on a subsequent input signal may also be configured to: derive a basis function extension set of the digital predistorter from the first basis function set, the basis function extension set comprising (d+1) groups of basis functions, wherein d represents the number of parameters of the operating conditions, wherein a first group of the basis function extension set comprises the first basis function set without any modification, and wherein each of the remaining d groups of the basis function extension set is determined as a transformation of the first basis function set according to a function of at least one of the parameters of the operating conditions. The digital predistorter may also be configured to weight the basis function extension set using the calculated digital predistortion parameters.

[0031] In some variations, another method for digitally predistorting a signal provided to a radio frequency (RF) transmit path is provided, the RF transmit path being configured to transmit radio signals in a plurality of sub-bands within a spectral range. The method includes: configuring a digital predistorter to predistort a signal including any spectral content within the spectral range, the configuration including: acquiring one or more data samples representing the operation of the RF transmit path for transmitting radio signals in different sub-bands, each data sample including a digital input signal provided to the RF transmit path for transmission and a digital sense signal representing a sense of a radio transmission in response to the provided digital signal, each digital input signal representing spectral content concentrated in a corresponding single corresponding sub-band within the spectral range; and updating parameters of the digital predistorter based on the acquired one or more data samples to mitigate nonlinear characteristics of the RF transmit path. The method also includes: receiving an additional input signal for transmission through the RF transmit path, the additional input signal representing spectral content in a particular sub-band within the spectral range; and using the configured predistorter to process the additional input signal to produce a predistorted signal for providing to the RF transmit path.

[0032] Embodiments of the additional methods may include at least some of the features described in the present disclosure, including any of the above features of the first method and the first linearization system and one or more of the following features.

[0033] The spectral range W may include n RF bands having bandwidths of approximately equal size.

[0034] Acquiring the one or more data samples may include controllably adjusting a wireless device including an RF transmit path to operate in one or more different subbands within the spectral range and measuring at least one sample set at each of the one or more different subbands.

[0035] Acquiring one or more data samples may include, during runtime, after initially calculating parameters of the digital predistorter, measuring additional sets of samples at additional subbands within the spectral range, and re-updating the parameters of the digital predistorter further based on at least some of the additional sets of samples.

[0036] The digital predistorter may be implemented in a user equipment (UE), and the plurality of subbands may be contiguously allocated bands for transmission.

[0037] The updated parameters of the digital predistorter can be configured to weight a basis function for predistorting another input signal, wherein the basis function includes complex terms depending on a τ-dimensional complex vector q corresponding to complex input samples of the digital distorter, and real terms depending on a d-dimensional real vector p representing values ​​of operating conditions associated with the operation of the RF transmit path, wherein d represents the number of parameters of the operating conditions associated with the operation of the RF transmit path.

[0038] The total number of basis functions used to digitally predistort the input signal may be proportional to the product of the number of terms in the first basis function set, b, and d+1, such that the total number of basis functions is proportional to b·(d+1).

[0039] Acquiring one or more data samples representing the operation of the RF transmit path in different sub-bands may also include: collecting corresponding operating condition parameters for the one or more data samples, wherein the corresponding operating condition parameters include one or more of the following: for example, a corresponding temperature of the RF transmit path, a corresponding average power, a corresponding supply voltage (V DD ), a corresponding voltage standing wave ratio (VSWR) associated with the RF transmit path, the age of each sample in the collected sample set, and / or corresponding channel information for collecting one or more data samples.

[0040] Processing the further input signal using the configured predistorter may include predistorting the further input signal according to:

[0041] in, Where v[t] is the output of the DPD at time t, u[t] is the input sample of the DPD at time t, and B k is the basis function for k=1 to n, q u [t] is the stack of input samples, p[t] is the d-dimensional real vector of operating conditions at time t, and x k are the calculated digital predistortion parameters used to weight the basis functions.

[0042] Updating the parameters of the digital predistorter may include deriving the digital predistortion parameters x according to an optimization process based at least in part on the acquired one or more data samples measured in different subbands. k , based on the basis function B k Weighted.

[0043] According to the optimization process, the parameters x of the digital predistorter are derived. k This may include adaptively adjusting the digital predistortion parameter x according to the following by minimizing the regularized mean square error function E(x): k:

[0044]

[0045] where ρ>0 is the regularization factor, v[t] is the output of the digital predistorter at time t, and y[t] is the output of the observation receiver coupled to the output of the RF transmit path.

[0046] Adaptively adjust the digital predistortion parameter x k This may include deriving the correlation matrices G and L according to the following when n<<100, where n represents the number of basis functions:

[0047]

[0048] in, as well as

[0049] The digital predistortion parameters x are derived according to k :

[0050]

[0051] Adaptively adjust the digital predistortion parameter x k may include approximating the digital predistortion parameters x according to a stochastic gradient process k .

[0052] Obtaining one or more data samples representing operation of an RF transmit path for transmitting radio signals in different sub-bands may include selecting one or more data samples from a plurality of data samples based on a diversity criterion calculated based on a diversity measure for each data record, wherein the plurality of records are arranged in a diversity database structure in which each record corresponds to one data sample of the one or more data samples.

[0053] In some variations, an additional linearization system for performing digital predistortion on a signal provided to a radio frequency (RF) transmit path is provided, the radio frequency (RF) transmit path being configured to transmit radio signals in a plurality of sub-bands within a spectral range. The system includes: an RF transmit path including at least one power amplifier producing an output having nonlinear distortion; and a digital predistorter including an adaptive portion and a compensator. The digital predistorter is configured to configure the compensator to predistort a signal including any spectral content within the spectral range, including: acquiring one or more data samples representing operation of the RF transmit path for transmitting radio signals in different sub-bands, each data sample including a digital input signal provided to the RF transmit path for transmission and a digital sense signal representing a sense of a radio transmission in response to the provided digital signal, each digital input signal representing spectral content concentrated in a corresponding single corresponding sub-band within the spectral range. Configuring the compensator also includes updating parameters of the digital predistorter based on the acquired one or more data samples to mitigate the nonlinear characteristics of the RF transmit path. The digital predistorter is additionally configured to: receive a further input signal for transmission via an RF transmit path, the further input signal representing spectral content in a particular sub-band within the spectral range; and process the further input signal using the configured compensator to produce a predistortion signal for provision to the RF transmit path.

[0054] Embodiments of the additional linearization system may include at least some of the features described in the present disclosure, including any of the above features for the various methods and for the first linearization system.

[0055] The digital predistorter configured to acquire one or more data samples may be further configured to controllably adjust a wireless device including an RF transmit path to operate in one or more different subbands within a spectral range, and measure at least one sample set at each of the one or more different subbands.

[0056] A digital predistorter configured to obtain one or more data samples may be further configured to, during runtime, after initially calculating parameters of the digital predistorter, measure additional sample sets at additional subbands within the spectral range, and re-update the parameters of the digital predistorter based also on at least some of the additional sample sets.

[0057] The digital predistorter configured to use the compensator may also be configured to: derive a basis function extension set of the digital predistorter from the first basis function set, the basis function extension set comprising (d+1) groups of basis functions, wherein d represents the number of parameters of the operating conditions, wherein a first group of the basis function extension set comprises the first basis function set without any modification, and wherein each of the remaining d groups of the basis function extension set is determined as a transformation of the first basis function set according to a function of at least one of the parameters of the operating conditions. The digital predistorter may also be configured to weight the basis function extension set with the updated digital predistortion parameters to process additional input signals.

[0058] A digital predistorter configured to obtain one or more data samples representing operation of an RF transmit path for transmitting radio signals in different subbands may also be configured to select one or more data samples from a plurality of data samples based on a diversity criterion calculated based on a diversity measure for each data record, wherein the plurality of records are arranged in a diversity database structure in which each record corresponds to one data sample of the one or more data samples.

[0059] In some variations, an additional method for digital predistortion is provided, the additional method comprising: obtaining a first set of digital predistortion (DPD) basis functions for controlling operation of a digital predistorter of a wireless device, the wireless device operating on at least one received input signal, the at least one input signal directed to a power amplification system, the power amplification system including a transmit chain having at least one power amplifier that produces an output having nonlinear distortion; and determining an adaptation set of DPD basis functions derived from the first set of DPD basis functions, the adaptation set of DPD basis functions dependent on operating conditions associated with operation of the wireless device, for controlling operation of the digital predistorter. The method also includes configuring the digital predistorter with DPD coefficients, wherein the DPD coefficients are determined for the adaptation set of DPD basis functions based at least in part on the transmit chain in response to observed samples of the at least one input signal.

[0060] Embodiments of further methods may include at least some of the features described in the present disclosure, including any of the above features of the first and second methods and the first and second linearization systems and one or more of the following features.

[0061] The basis function includes complex terms that depend on a τ-dimensional complex vector q corresponding to complex input samples of the digital distorter, and real terms that depend on a d-dimensional real vector p representing values ​​of operating conditions associated with the operation of the wireless device on at least one received input signal, where d represents the number of parameters of the operating conditions associated with the operation of the wireless device.

[0062] The total number of basis functions used to digitally predistort at least one input signal may be proportional to the product of the number of terms b in the first basis function set and d+1, such that the total number of basis functions is proportional to b·(d+1).

[0063] The basis functions may include (d+1) groups of basis functions, wherein a first group of basis functions includes a first set of basis functions without any modification, and wherein each of the remaining d groups of basis functions is determined as a product of the first set of basis functions and one of the operating conditions associated with the wireless device.

[0064] Configuring the predistorter using the DPD coefficients may include generating the DPD coefficients x according to k The DPD coefficients are calculated by an optimization process as weights of the basis functions that minimize an error function based on observed samples of the transmission chain and samples of at least one input signal.

[0065] Calculate the DPD coefficient x according to the optimization process k This may include adaptively adjusting the DPD coefficient x according to the following minimization of the regularized mean square error function E(x): k :

[0066]

[0067] in, Where ρ>0 is the regularization factor, B k is the basis function for k=1 to n, q v [t] is the stack of input samples at time t, p[t] is a d-dimensional real vector of operating conditions at time t, v[t] is the output of the digital predistortion (DPD) compensator at time t, and y[t] is the output of the observation receiver coupled to the output of the transmit chain.

[0068] Adaptively adjust the DPD coefficient x k This may include deriving the correlation matrices G and L according to the following when n<<100, where n represents the number of basis functions:

[0069]

[0070] in,

[0071] as well as

[0072] The DPD coefficient x is derived from k :

[0073]

[0074] Adaptively adjust the DPD coefficient x kThis may involve approximating the DPD coefficients x according to a stochastic gradient process k .

[0075] The method may further include collecting a data record representing a sample of the at least one input signal and a corresponding one or more observation samples of the transmission chain in response to the sample of the at least one input signal.

[0076] Collecting the data records may include collecting the data records in one or more spectral channels, the one or more spectral channels being located within a wideband spectral range, wherein each of the one or more spectral channels has a respective bandwidth that is less than a range bandwidth of the spectral range.

[0077] In some variations, an additional linearization system for performing digital predistortion is provided, the additional linearization system comprising: a transmission chain including at least one power amplifier that produces an output having nonlinear distortion; and a digital predistorter including an adaptive portion and a compensator. The digital predistorter is configured to obtain a first set of digital predistortion (DPD) basis functions for controlling operation of a digital predistorter of a wireless device, the wireless device operating on at least one received input signal; and determine an adaptation set of DPD basis functions derived from the first set of DPD basis functions, the adaptation set of DPD basis functions depending on operating conditions associated with operation of the wireless device, for controlling operation of the digital predistorter. The digital predistorter is further configured to configure the compensator using DPD coefficients, wherein the DPD coefficients are determined at least in part based on the transmission chain in response to observed samples of the at least one input signal for the adaptation set of DPD basis functions.

[0078] Embodiments of additional linearization systems may include at least some of the features described in this disclosure, including any of the above features for various methods and for various linearization systems and one or more of the following features.

[0079] The basis functions may include (d+1) groups of basis functions, wherein a first group of basis functions includes a first set of basis functions without any modification, and wherein each of the remaining d groups of basis functions is determined as a transformation of the first set of basis functions based on a function of at least one of the parameters of the operating conditions.

[0080] Each of the remaining d groups of basis functions may be determined as a product of the first set of basis functions and one of the operating conditions associated with the wireless device.

[0081] The digital predistorter configured to configure the compensator using the DPD coefficients may also be configured to generate the DPD coefficients x according to kThe DPD coefficients are calculated by an optimization process as weights of the basis functions that minimize an error function based on observed samples of the transmission chain and samples of at least one input signal.

[0082] The digital predistorter can also be configured to collect data records in one or more spectral channels, which data records represent samples of at least one input signal and corresponding one or more observation samples of the transmit chain in response to the samples of the at least one input signal, and the one or more spectral channels are located within a wideband spectral range, wherein each of the one or more spectral channels has a corresponding bandwidth that is smaller than the range bandwidth of the spectral range.

[0083] In some variations, yet another method for digital predistortion is provided, the yet another method comprising: collecting data for a wireless device including a transmit chain including at least one power amplifier that produces an output having nonlinear distortion. The data comprises data records, wherein each data record comprises input signal data and corresponding output signal data for the transmit chain and is also associated with a set of operating conditions, wherein each data record is associated with a diversity measure representing a degree of distinguishability of the corresponding data record from other collected data records, and wherein collecting the data comprises collecting the data in one or more spectral channels, the one or more spectral channels being within a wideband spectral range, wherein each of the one or more spectral channels has a corresponding bandwidth that is less than a range bandwidth of the spectral range. The method also includes: selecting at least some of the data records according to a diversity criterion calculated based on a diversity measure for each of the data records; and calculating digital predistortion parameters for controlling a digital predistorter of the wireless device based on the selected at least some of the data records, wherein calculating the digital predistortion parameters includes: obtaining a first set of digital predistortion (DPD) basis functions for controlling the operation of the digital predistorter; determining an adaptation set of DPD basis functions derived from the first set of DPD basis functions, the adaptation set of the DPD basis functions depending on operating condition values ​​corresponding to the selected at least some of the data records; configuring the digital predistorter using the digital predistortion parameters, the digital predistortion parameters being determined for the adaptation set of the DPD basis functions based at least in part on the selected at least some of the data records. The method also includes applying the digital predistorter configured using the calculated digital predistortion parameters to a subsequent input signal.

[0084] Embodiments of yet another method may include at least some of the features described in this disclosure, including any of the above features for the various methods and for the various linearization systems.

[0085] In some variations, systems are provided that are configured to perform the steps of one or more of the methods provided above.

[0086] In some variations, a design structure encoded on a non-transitory machine-readable medium is provided, wherein the design structure includes elements that, when processed in a computer-aided design system, generate a machine-executable representation of one or more of a system, a digital predistorter, a transmit chain, and / or any corresponding modules thereof as described herein.

[0087] In some variations, an integrated circuit definition data set is provided that, when processed in an integrated circuit manufacturing system, configures the integrated circuit manufacturing system to manufacture one or more of a system, digital predistorter, transmit chain, and / or any corresponding modules thereof as described herein.

[0088] In some variations, a non-transitory computer-readable medium is provided, the non-transitory computer-readable medium being programmed with a set of computer instructions executable on a processor, which when executed results in operations including the steps of the various methods described above.

[0089] Other features and advantages of the invention will be apparent from the following description, and from the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] These and other aspects will now be described in detail with respect to the following figures.

[0091] Figure 1 is a block diagram of a linearization system configured to implement various DPD adaptation methods as described herein.

[0092] Figure 2 is a diagram illustrating the collection of samples of narrow sub-bands from a wide band spectral range.

[0093] Figure 3 Included are graphs comparing performance results of DPD compensation systems implemented using two different methods, namely a wideband sample collection method and a narrow sub-band sample collection method.

[0094] Figure 4 is a flow chart of an example process for digital predistortion of a radio frequency (RF) transmit path configured to transmit radio signals in a plurality of sub-bands within a spectral range.

[0095] Figure 5 is a flow chart of an example process for digital predistortion using adaptive basis functions.

[0096] Fig. 6A Included are figures illustrating simulation results of an optimization process performed without ensuring a well-diverse sample.

[0097] Figure 6BIncluded are graphs showing simulation results when the optimization process uses samples that are subjected to a diversification-based distribution (randomization) method to promote sample diversity.

[0098] Figure 7 is a flow chart of an example process for implementing digital predistortion using a DPD adaptation process according to a diversity-based approach.

[0099] Like reference numbers in the various drawings indicate like elements. DETAILED DESCRIPTION

[0100] In very general terms, the following includes a description of a number of techniques that may be used individually or, preferably, in combination to provide a digital predistorter that may provide high performance over a wide range of operating conditions and that is capable of responding to rapid changes between such operating conditions.

[0101] The techniques are particularly applicable to radio frequency (RF) user equipment (UE), such as a cellular telephone, which is controlled (e.g., by a base station (BS)) to take advantage of rapid changes, for example, in spectral position (e.g., transmit channel or subcarrier), power level, or output load. However, it should be understood that the techniques are not limited to implementation on user equipment and that the techniques may also be used on base stations. In addition, although described in the context of predistortion for over-the-air RF applications, the techniques may be applied to other areas, including coaxial cable communications (e.g., for use in DOCSIS standard modems), audio communications (e.g., for linearizing audio, ultrasound, ultra-low frequencies), and optical communications (e.g., to mitigate nonlinear characteristics of optical communication paths).

[0102] The first of these techniques is motivated in part by the varying spectral positions of the transmit spectrum. In this technique, generally, the predistorter is configured to process input signals that span relatively narrow subbands within a wider available bandwidth, and it is largely or completely unaware of the location and / or width of the specific subbands that the transmitter uses to operate at any given time, in the sense that the predistorter is "generic". It will be appreciated that one challenge with this technique is that the feedback available for updating the predistorter parameters senses characteristics over only a small portion of the entire available bandwidth at any one time, in the sense that the data available for adapting the predistorter is sparse.

[0103] The second of these techniques explicitly accounts for changes in certain aspects of the operating conditions by introducing parameters that represent those aspects and incorporating dependencies on those parameters into the model structure of the predistorter. Using this approach, a change in one of these aspects involves changing the value of the corresponding parameter of the predistorter model without having to wait for a slower adaptive process to react to changes in that aspect of the operating conditions. As an example, the dependency on transmit power can be parameterized and the predistorter can react to power changes virtually immediately. In another example, the basis function used to predistort the input signal, which is generally a function of the expected input, can also be adapted based on changes in operating conditions.

[0104] In an example implementation of a second technique, a method for digital predistortion is provided, the method comprising obtaining a first set of digital predistortion (DPD) basis functions for controlling operation of a digital predistorter of a wireless device, the wireless device operating on at least one input signal directed to a power amplification system including a transmit chain having at least one power amplifier producing an output having nonlinear distortion. The method additionally comprises: determining an adaptation set of DPD basis functions derived from the first set of DPD basis functions, the adaptation set of DPD basis functions depending on operating conditions associated with operation of the wireless device (e.g., including environmental parameters) for controlling operation of the digital predistorter; and configuring the digital predistorter with DPD coefficients, the DPD coefficients being determined at least in part based on the transmit chain responding to observed samples of the at least one input signal for the adaptation set of DPD basis functions.

[0105] A third of these techniques addresses the possibility that during operation, a range of operating conditions are not used in a representative manner, or follow a temporal pattern that may cause adaptation of the predistorter to reduce its ability to respond to new operating conditions, which adaptation may not be adequately represented in terms of its effect on the configuration of the predistorter. Very generally, the technique uses a sampling of operating data based on which the configuration is adapted, which maintains sufficient diversity over possible operating conditions to provide good performance immediately after a change from one configuration to another.

[0106] In an example implementation of the third type of technology, digital predistortion adaptation is implemented using data samples selected according to a diversity metric. In some such implementations, a method for digital predistortion is provided, the method comprising collecting data for a wireless device including a transmission chain, the transmission chain including at least one power amplifier that produces an output with nonlinear distortion. The collected data includes data records, wherein each data record includes input signal data and corresponding output signal data for the transmission chain and is also associated with a set of operating condition values. Each data record is associated with a diversity metric representing the degree of distinguishability of the corresponding data record from other collected data records. Examples of such a diversity metric include a distance metric between elements of different records (large distances generally indicate high distinguishability between such different records), wherein an example of such a distance calculation is the maximum absolute value of the difference between corresponding elements of the data records. The method additionally includes: selecting at least some of the data records according to a diversity criterion calculated based on the diversity metric of each data record; calculating digital predistortion parameters for controlling a digital predistorter of the wireless device based on the selected at least some of the data records; and applying the digital predistorter configured with the calculated digital predistortion parameters to a subsequent input signal.

[0107] Below Figure 1 The techniques summarized above are described in the context of a predistortion system 100 shown in block diagram form in FIG. Figure 1, u is an input signal representing a desired signal to be transmitted via the transmission part F 120. The signal u comprises a sequence of complex I / Q samples ..., u[t-1], u[t], u[t+1], ... provided to the system 100 in digital form. As introduced above, this input signal generally occupies a relatively small portion of the total available bandwidth of the system. However, by implementing an adaptive process using samples collected on a large number of various bands / sub-bands within the available bandwidth of the system (e.g., the frequency range of the available communication channel set) (i.e., the implementation is referred to as a first technique, as discussed in further detail below), the adaptive DPD compensation system 100 can be implemented as a universal DPD compensator that is configured to operate at different spectral bands (e.g., within a wide spectrum range W, the wide spectrum range W includes bands operated by different corresponding commercial cellular operators, different available commercial bands, different cellular technologies, etc.). The technology is also suitable for use with similar systems described in the following: PCT application No. PCT / US2019 / 031714, entitled “Digital Compensation for a Non-Linear System,” filed on May 10, 2019, U.S. application No. 16 / 004,594, entitled “Linearization System,” filed on June 11, 2018, and U.S. application No. 16 / 004,713, entitled “Multi-Band Linearization System,” filed on June 11, 2018, each of which is incorporated herein by reference.

[0108] The compensator C 110 is electrically coupled to a transmit section F 120, which represents a signal chain from a digital signal (at baseband or possibly at an intermediate band) to a radio frequency signal that is provided to or emitted from an antenna, and includes components such as modulators, power amplifiers, analog transmit lines, etc., which can introduce nonlinear characteristics into the signal path. The transmit section F 120 (also referred to as a transmit chain or transmit chain) implements a digital-to-analog subsystem that includes, for example, a DAC / PA combination. In a wireless implementation, the transmit section F 120 also includes a modulator (e.g., implemented using a local oscillator) to allow the resulting analog signal to be transmitted within a certain RF band within the spectrum range W in which the system 100 is configured to operate. The specific band can be selected and / or adjusted (e.g., during a switching operation) based on the availability of cellular resources (e.g., available RF communication bands).

[0109] The compensator (also referred to as a digital predistorter) C 110 is configured to compensate for the nonlinear characteristics of the transmit portion F 120 (such nonlinear behavior may be caused by the operation of one or more of the components of the transmit portion F 120 (such as a power amplifier, DCA, modulator, etc.)). Specifically, the predistorter receives a desired signal u and outputs a corresponding predistorted signal v, which is a function of u, where this function is parameterized by a configuration parameter x. The sensing portion S130 (also referred to as an observation path) senses the output of the transmit portion F 120 and provides a sensed signal y in digital form. Ideally, if the predistorter 110 is able to accurately invert all nonlinear effects of the transmit portion, the sensed signal y is the same as the desired input u (or a scaled version, such as an amplified version). The sensing portion S130 implements an analog-to-digital subsystem, and the sensing portion S130 represents a real-time output measure (e.g., a demodulator / ADC combination) of the output w of the transmit portion F120. Block S130 produces observed digital samples represented as signal y. In some embodiments, at least one batch of observed samples is stored in a storage device (eg, a buffer or a group of buffers) in electrical communication with at least one subsystem of the system 100 .

[0110] The adaptive portion A 140 (also referred to as an adapter) receives the desired signal u and the sensed signal y and determines the predistortion parameters x to be used so as to approximate the ideal case where y is the same as u. As will be discussed in more detail below, in some example embodiments, the system 100 is configured to derive and apply a predistortion process that uses adaptive basis functions based on operating conditions (e.g., environmental characteristics) of the system 100. In at least some combinations of techniques that explicitly model aspects of the operating conditions, the adaptive portion receives operating parameters p (in Figure 1 1 and 2). Such sensing devices may include, for example, temperature sensors, current and / or voltage sensors, RF power meters, circuit and / or processor-based devices that derive operating condition parameters (such as voltage standing wave ratio or VSWR, values ​​indicative of the age of the record (such as a timestamp), etc.

[0111] More specifically, samples of the intermediate signal v and the observed samples y are provided to the adaptive portion 140A, which implements an adaptive process (e.g., an optimization process, a coefficient selection process that selects predetermined coefficients from a lookup table, a coefficient interpolation process, etc.) based on the data obtained for v and y (and in some cases u) to derive a vector x of weighting coefficients for the basis functions applied by the compensator C 110. The observed samples y as well as the intermediate samples v and / or the input samples u may all be stored in a buffer (not shown, which may be part of any subsystem of the system 100). The contents of the buffer (which may be arranged as a matrix or according to some other suitable data structure) may be occasionally updated when new samples that are deemed more useful to the adaptive process (because they are more recent and / or meet certain data criteria such as data uniformity) become available.

[0112] In at least some implementations, the compensator (predistorter) C 110 implements a nonlinear transformation, which can be expressed as:

[0113]

[0114] In this equation, q u [t] is a window of τ past input values,

[0115]

[0116] and p[t] is a vector (or scalar) representing various aspects of the current operating condition. Note that in versions where such aspects are not explicitly modeled, the dependency on p[t] can be ignored. Figure 1 The parameter x shown in includes n complex-valued parameters x k Vector.

[0117] Item B k (q, p) is called a "basis function" that can output a scalar complex-valued function of the inputs q and p. A variety of such functions can be used, for example, as described in PCT Application No. PCT / US2019 / 031714, entitled "Digital Compensation for a Non-Linear System," filed on May 10, 2019. Note that q u The number of samples in depends on the basis function used, depending on the maximum lag τ used by any basis function. Very generally, in some embodiments, the basis function B may be selected during operation according to the operating conditions. k, and when new data samples v and y are made available to the adaptation portion A 140 (e.g., in an incremental update process), the adaptation portion A 140 modifies (i.e., adapts) the parameters x (where the parameters x weight the basis functions). One or more such processes are described later in this document. Thus, the adaptive DPD compensation system 100 can be configured to adaptively determine the DPD coefficients (e.g., selected from a lookup table / matrix containing predetermined DPD coefficients associated with different operating points, or dynamically derived based on the current sample and previously obtained samples).

[0118] As noted, in some example approaches, the basis functions may be adaptive functions that vary according to operating conditions, thus providing a greater number and variety of basis functions in addition to an initial set of carefully pre-selected basis functions. Thus, in some examples, the adaptability of the basis functions used for the predistortion operation may be made dependent on a vector p representing d operating conditions, which is introduced into the predistortion model by essentially introducing a linear dependency on each of these terms. This may be accomplished by using x 3(k-1)+1 B k0 (q)+x 3(k-1)+2 B k0 (q)p1+...+x 3(k-2)+d+1 B k0 (q)pd replaces x k B k (q,p) is used to represent it.

[0119] Therefore, the number term in the adaptation parameter x increases by a factor d+1. In this example, the complexity of the basis function remains unchanged, because the same basis function B k0 for a set of (d+1) terms. In other implementations, a different basis function may be used for each aspect of the operating condition, such that the basis functions may depend in part on the values ​​of concurrent environmental parameters measured / determined during the adaptive process, where the number of total basis functions used to digitally predistort the input samples depends on the number of operating conditions used. For example, if two environmental parameters (e.g., temperature and VDD) are measured for the collected samples, the number of basis functions used for the digital predistortion process will be three (3) times the number of basis functions used in an implementation that does not incorporate environmental parameters (or other types of operating conditions) into the adaptive process.

[0120] As noted above and as will be discussed in more detail below, the system 100 may also be configured to be more robust and more effectively adaptive through the implementation of a diverse matrix adaptation approach (referred to above as the "third technique"). In such an approach, past samples (q u [t],u[t]-y[t],p[t]) or for some adaptive processes (q y[t],v[t]-y[t],p[t]), are not used in a uniform distribution over past time points. One implementation of such non-uniform sampling is to maintain a buffer V of a subset of past samples. (q (i) ,v (i) ,p (i) ) samples are maintained by the system to represent a diversity of past operating conditions without over-representing any particular condition. Generally, when a new time sample (q[t], v[t], p[t]) is available for the adaptation portion, the adaptation portion A 140 determines whether to keep this time sample in the buffer V, and if so, determines which current entry in the buffer to pop.

[0121] A general method for determining whether to eject the current sample and replace it with a new time sample is to do so if such replacement can increase the diversity or uniformity of the sampling of operating conditions in the buffer. For example, for each data i in the buffer, an easily computable distance between data samples is defined as the maximum absolute difference between the components of the data sample. In general, it is desirable to have the data points in the buffer have as large a minimum distance between data points as possible in order to cover a larger hyperspace, thereby promoting diversity throughout the buffer. The new time sample similarly has a minimum distance h to the data points in the buffer. i Then pop up the minimum value min(r i ,h i ) and replaces them with new samples. In the adaptive process, the updated buffer V is used to update the predistortion parameters x.

[0122] As noted, the adaptive portion A 140 is configured to compute DPD coefficients that weight the basis functions applied to the input signal u[t] by the compensator C 110. In at least some examples, the parameters x of the predistorter are updated using an iterative update process (e.g., using a stochastic gradient process). In other examples, the parameters x are updated in a batch process (e.g., using a Gramian matrix inversion process).

[0123] In the iterative process, the goal is to optimize (i.e., minimize) the objective

[0124]

[0125] Among them, e[t]=v[t]-y[t],

[0126] The sum is over past time samples, or if a buffer V is used, over the time samples represented in the buffer.

[0127] In the stochastic gradient method, a uniformly randomly chosen time τ is chosen and the current independent variable value x is updated to

[0128]

[0129] in,

[0130]

[0131] as well as

[0132] a(τ)=[B1(q y [t],p[t]),...,B N (q y [t],p[t])]

[0133] The step size α may be adjusted (eg, decreased) over time. Further details regarding implementations of various optimization processes and one or more agile adaptation techniques described herein that may be used in conjunction with the system 100 are provided below.

[0134] Return to Figure 1 , Figure 1 The system 100 described in the foregoing may be at least a portion of a digital front end of a device or system (such as, for example, a network node of a WWAN base station or a WLAN access point, or belonging to a mobile device). As such, the system 100 may be electrically coupled to a communication module that enables wired communication with a remote device (e.g., via a network interface for a wired network connection such as through Ethernet), or enables wireless communication with such a remote device. In some embodiments, the system 100 may be used locally within a device or system to process various locally generated signals (pre-distortion processing or otherwise) to generate a resulting processed signal (whether or not the resulting signal is then transmitted to a remote device).

[0135] In some embodiments, at least some of the functions of the linearization system 100 may be implemented using a controller (e.g., a processor-based controller) included with one or more modules of the system 100 (compensator C 110, adaptive portion A 140, or any other module of the system 100). The controller may be operably coupled to the various modules or units of the system 100 and configured to calculate approximations of the gradient of the error function (as will be discussed in more detail below), update the DPD coefficients based on those approximations, and perform digital predistortion using the continuously updated DPD coefficients. The controller may include one or more microprocessors, microcontrollers, and / or digital signal processors that provide processing functions and other computing and control functions. The controller may also include dedicated logic circuitry, such as an FPGA (field programmable gate array), an ASIC (application specific integrated circuit), a DSP processor, a graphics processing unit (GPU), an accelerated processing unit (APU), an application processor, a customized dedicated circuit, etc. to at least partially implement the processes and functions for the system 100. The controller may also include a memory for storing data and storing software instructions for executing programmed functions within the device. Generally speaking, computer-accessible storage media may include any non-transitory storage media that a computer can access during use to provide instructions and / or data to the computer. For example, computer-accessible storage media may include storage media such as magnetic or optical disks and semiconductor (solid-state) memory, DRAM, SRAM, etc.

[0136] The outlined method can be used alone or in combination with any other DPD adaptation method discussed herein. Thus, for example, embodiments of the methods and techniques described herein include a method for digital predistortion, the method comprising collecting data for a wireless device including a transmission chain, the transmission chain including at least one power amplifier that produces an output with nonlinear distortion, the data comprising data records, wherein each data record includes input signal data and corresponding output signal data for the transmission chain and is also associated with a set of operating conditions. Each data record is associated with a diversity metric representing a degree of distinguishability of the corresponding data record from other collected data records. Collecting the data includes collecting data in one or more spectral channels, the one or more spectral channels being located within a wideband spectral range, wherein each of the one or more spectral channels has a corresponding bandwidth that is less than a range bandwidth of the spectral range. The method also includes: selecting at least some of the data records based on a diversity criterion calculated based on the diversity metric of each data record; and calculating digital predistortion parameters for controlling a digital predistorter of the wireless device based on at least some of the selected data records. Calculating digital predistortion parameters includes: obtaining a first set of digital predistortion (DPD) basis functions for controlling operation of a digital predistorter; determining an adaptation set of DPD basis functions derived from the first set of DPD basis functions, the adaptation set of DPD basis functions depending on operating condition values ​​corresponding to at least some selected data records (e.g., an average of corresponding operating condition values ​​from each selected record may be calculated); and configuring the digital predistorter using the digital predistortion parameters, the digital predistortion system being determined for the adaptation set of DPD basis functions based at least in part on the selected at least some data records. The method additionally includes applying the digital predistorter configured using the calculated digital predistortion parameters to a subsequent input signal.

[0137] Additional details about the various agile adaptive techniques described in this paper are as follows.

[0138] First technology: Wideband adaptation based on narrowband measurement

[0139] As discussed herein, agile estimation of digital predistortion (DPD) parameters may be implemented with a general parameterized digital predistorter operable over a wideband spectral range W, which may be based on samples collected over a portion of the wideband spectral range at different time intervals (e.g., based on samples collected in narrow sub-bands of the spectral range). Such samples may be collected using an adjustable narrowband wireless communication device configured to operate in different sub-bands within the spectral range.

[0140] Thus, in some embodiments, a method for digital predistortion of a signal is disclosed, the signal being provided to a radio frequency (RF) transmit path (e.g., transmit section F 120) configured to transmit a radio signal in a plurality of sub-bands within a spectral range. The method includes configuring a digital predistorter (e.g., Figure 1 The configuration includes acquiring one or more data samples representing the operation of the RF transmit path to transmit radio signals in different sub-bands, wherein each data sample includes a digital input signal provided to the RF transmit path for transmission and a digital sensing (e.g., sensed by a sensing portion S130) signal representing sensing of the radio transmission in response to the provided digital signal. Each digital input signal represents spectral content concentrated in a corresponding single corresponding sub-band within the spectral range (e.g., concentrated or limited to a defined frequency range for a single communication channel or subcarrier). Configuring the digital predistorter also includes updating parameters of the digital predistorter (e.g., by an adaptive process generally implemented using an adaptive portion A 140) based on the acquired one or more data samples to mitigate nonlinear characteristics of the RF transmit path. The process also includes: receiving an additional input signal for transmission through the RF transmit path, the additional input signal representing spectral content in a specific sub-band within the spectral range; and using the configured predistorter to process the additional input signal to produce a predistorted signal for providing to the RF transmit path.

[0141] Therefore, in the implementation of the first technique, a universal parameterized predistorter is implemented that is robust to changing spectral conditions. That is, in the method of the first technique, a single predistorter implementation can be implemented that can operate on different subcarriers and does not have to adapt to new spectral conditions when the spectral characteristics of the channel (carrier frequency, bandwidth, communication protocol used, average output power, etc.) change (for example, because of moving to a different cell, changing to a different band within the same cellular area due to changes caused by base station allocation, etc.). Since the spectral characteristics may change frequently, the use of a robust universal digital predistorter configured to operate over the entire spectral range W can reduce the computational complexity that would result from having to adapt to each change in spectral operating conditions.

[0142] To implement a general parameterized digital predistorter, samples from sub-bands of a spectral range (i.e., bands within the spectral range W having a bandwidth narrower than the bandwidth W associated with the spectral range) are collected. Depending on the specific usage of the wireless device, the availability of spectral resources (e.g., whether some available bands are allocated to other wireless devices), and the existing environmental conditions, the collection of samples may be performed at design time by methodically traversing the spectral range, or may be collected during normal operation of the wireless device (including the digital predistorter) as the wireless device hops from one sub-band to another. When a wireless device (including a digital predistorter such as Figure 1 When a digital predistortion system of the system 100 dynamically collects samples during normal operation and occasionally (at regular or irregular intervals) updates DPD parameters, the collected samples and operating conditions (which may include spectral operating conditions) may be stored together in a diversity matrix such as the diversity matrix described herein (i.e., to collect samples that promote sample diversity). Occasionally, the digital predistortion system updates the DPD coefficients using the currently stored records. In such an example, the updated coefficients will reflect good diverse samples, including possible diverse samples collected from a variety of subbands. Alternatively, the wireless device may collect samples according to some other scheme for obtaining representative samples obtained for multiple subbands.

[0143] about Figure 2 , a diagram 200 is provided illustrating the collection of samples of narrow subbands (210, 212, and 214) from a wideband spectral range having a bandwidth of 100 MHz. In this figure, the wireless device jumps to (or systematically slides into) different subbands and collects samples from the narrowband subbands. In addition to the input samples (and / or observed samples), operating conditions that may include spectral information may also be collected and selectively stored (e.g., based on a calculated diversity metric, based on which a decision is made whether a particular sample should be stored or discarded). Occasionally, the adaptive process uses the narrowband samples at regular or irregular intervals to generate or update predistortion parameters, such as DPD coefficients weighted by basis functions (which may be adaptive basis functions, as described herein). In some embodiments, sample data collection may be performed by sampling the input (and / or observed output of the transmit chain) at the full bandwidth (i.e., at the Nyquist frequency for the spectral range W) or at a narrower bandwidth of the subband (optionally upsampling the collected samples, for example, during a DPD parameter update process).

[0144] Figure 3Figures 310 and 320 are included, which compare the performance results of DPD compensation systems implemented using two different methods. In the first implementation, corresponding to Figure 310, wideband signal samples are used to generate DPD coefficients (i.e., samples of the wideband signal are used to train the DPD system). In the second implementation, corresponding to Figure 320, samples collected from multiple narrowband sub-bands are used to generate DPD coefficients. In the implementation corresponding to Figures 310 and 320, the derivation of DPD coefficients is performed by accumulating and averaging new and old correlation matrices. As shown in both Figures 310 and 320, the DPD system trained using samples from multiple narrowband sub-bands achieves generally higher and more robust performance (based on adjacent channel leakage ratio or ACLR behavior for both implementations).

[0145] Next about Figure 4 , a flow chart of an example process 400 for digitally predistorting a signal provided to a radio frequency (RF) transmit path configured to transmit radio signals in a plurality of sub-bands within a spectral range is shown. The process 400 includes configuring 410 a digital predistorter (such as a Figure 1 The digital predistorter implemented by the system 100 of the present invention is used to predistort a signal including any spectral content within a spectral range. The configuration operation 410 of the process 400 includes obtaining 412 one or more data samples representing the operation of the RF transmit path to transmit a radio signal in different sub-bands. Each data sample includes a digital input signal (e.g., Figure 1 As shown in and also in Figure 4 u[t] and / or v[t] as described in the embodiment of the present invention) and a digital sensing signal representing the sensing of the radio transmission in response to the provided digital signal (e.g., Figure 1 and Figure 4 y[t] shown in ). Each digital input signal represents the spectral content in a corresponding single corresponding sub-band concentrated in the spectral range. This spectral information can also be used as Figure 1 Thus, in some embodiments, acquiring one or more data samples representing the operation of the RF transmit path in different sub-bands may further include: collecting corresponding operating condition parameters for the one or more data samples, wherein the corresponding operating condition parameters include one or more of the following, for example, a corresponding temperature of the RF transmit path, a corresponding average power, a corresponding supply voltage (V DD ), a corresponding voltage standing wave ratio (VSWR) associated with the RF transmit path, a respective age of the collected sample sets, and / or corresponding channel information for collecting one or more data samples. In some examples, the spectral range W includes n RF bands of approximately equal-sized bandwidths.

[0146] In some embodiments, acquiring one or more data samples may include controllably adjusting a wireless device including an RF transmit path to operate in one or more different sub-bands within a spectrum range (e.g., by controlling modulation / demodulation functions by adjusting a local oscillator, adjusting bandpass filtering operations, etc.), and measuring at least one sample set at each of the one or more different sub-bands. Acquiring one or more data samples may include: during runtime, after initially calculating parameters of a digital predistorter, measuring additional sample sets at additional sub-bands within the spectrum range, and also updating the parameters of the digital predistorter based on at least some of the additional sample sets. In some embodiments, acquiring one or more data samples representing operation of the RF transmit path to transmit radio signals in different sub-bands may include: selecting one or more data samples from a plurality of stored data samples according to a diversity criterion calculated based on a diversity metric for each data record, wherein the plurality of records are arranged in a diversity database structure, wherein each record in the diversity database structure corresponds to one of the one or more data samples.

[0147] The configuration operation 410 also includes updating 414 parameters of the digital predistorter (e.g., DPD coefficients weighting the basis functions) based on the acquired one or more data samples to mitigate nonlinear characteristics of the RF transmit path. For example, in some embodiments, the basis functions are adaptive basis functions (the basis functions may be extended from a basis set based on, for example, operating conditions) that include complex entries dependent on a τ-dimensional complex vector q corresponding to complex input samples of the digital distorter, and real entries dependent on a d-dimensional real vector p representing values ​​of operating conditions associated with operation of the RF transmit path. d represents the number of parameters of the operating conditions associated with operation of the RF transmit path. In such embodiments, the total number of basis functions used to digitally predistort the input signal is proportional to the product of the number of entries b in the first basis function set (i.e., the basis set) and d+1, such that the total number of basis functions is proportional to b·(d+1).

[0148] Continue About Figure 4 , process 400 also includes receiving 420 an additional input signal (at Figure 4 The signal u subsequent [t]), wherein the additional input signal represents spectral content in a particular sub-band within the spectral range. Process 400 additionally includes processing the additional input signal using 430 a predistorter configured according to operation 410 to generate a predistorted signal for providing to the RF transmit path (at Figure 4 The signal v subsequent [t]).

[0149] In some examples, processing the further input signal using the configured predistorter may include predistorting the further input signal according to:

[0150] in, Where v[t] is the output of the DPD at time t, u[t] is the input sample of the DPD at time t, and B k is the basis function for k=1 to n, q u [t] is the stack of input samples, p[t] is the d-dimensional real vector of operating conditions at time t, and x k are the digital predistortion parameters calculated to weight the basis functions.

[0151] Updating the parameters of the digital predistorter may include deriving the digital predistortion parameters x according to an optimization process based at least in part on one or more data samples measured in different subbands obtained. k , based on the basis function B k In some examples, the parameters x of the digital predistorter are derived according to the optimization process. k This may include adaptively adjusting the digital predistortion parameter x according to the following by minimizing the regularized mean square error function E(x): k :

[0152]

[0153] where ρ>0 is the regularization factor, v[t] is the output of the digital predistorter at time t, and y[t] is the output of the observation receiver coupled to the output of the RF transmit path.

[0154] Adaptively adjust the digital predistortion parameter x k This may include, for example, deriving the correlation matrices G and L according to the following when n<<100, where n represents the number of basis functions:

[0155]

[0156] in,

[0157] The digital predistortion parameters x are derived according to k :

[0158]

[0159] In the above expression, x is the complex conjugate x k Column vector of .

[0160] In some embodiments, the digital predistortion parameter x is adaptively adjusted. k may include approximating the digital predistortion parameters x according to a stochastic gradient processk .

[0161] Second technique: parameterized digital predistortion basis functions

[0162] As noted, the coefficients derived by the adaptive portion A 140 are used to weight the basis functions used by the compensator (predistorter) C 110 to predistort the input u. In some embodiments, predistorting the signal u by the compensator C 110 requires minimizing the difference between y (the output of the sensing portion S130) and u (the input sample) as small as possible. In order to add agility and versatility to the predistortion operation, some example embodiments described herein utilize an adaptive set of basis functions to construct an extended set of basis functions based on concurrent operating conditions. In other words, when operating conditions (e.g., environmental characteristics) change, the specific set of basis functions operating on the input samples is adjusted according to the existing operating conditions.

[0163] An example method of using an adapted set of basis functions that is dependent on operating conditions (to achieve agile estimation of DPD parameters, improve response speed to changing conditions and increase robustness of the DPD system) is based on deriving an adapted set of basis functions from an initial / first set of DPD basis functions. The first set of DPD functions may be a predetermined, carefully selected set of functions B k (·), function set B k (·) depends on the input samples of the DPD system (and is configured to provide, for example, good DPD performance for a particular transmit chain or wireless device), but does not depend on operating conditions. Therefore, the initial (basic) set of basis functions (from which the derivative / derivative set of basis functions can be generated) includes terms that apply to the input samples of the DPD system (i.e., u[tk], ..., u[t-1], u[t], u[t+1], ...). The initial set of functions can then be used to construct accompanying (child or derivative) basis functions, with added terms that are functions of the operating conditions (e.g., parameters representing environmental characteristics). Therefore, as more terms that depend on operating conditions are added to the initial (basic) set of basis functions, the size of the adaptive set of basis functions increases.

[0164] Given by the following ( Figure 1 The common representation of the compensator C 110 is:

[0165]

[0166] in, is a stack of recent (near t) baseband input samples. As noted above, the vector p[t] is a d-dimensional real vector of operating conditions at time t, x k (k=1,...,n) is a complex n-dimensional vector x∈C n (compensator coefficients) elements, B k(q, p) are carefully selected “basis functions” of a τ-dimensional complex vector q and a d-dimensional real vector p (“careful” selection means that the operation of computing a single sample v[t] of the corrected signal is significantly less than n scalar multiplications; examples of carefully selected basis functions are provided in: PCT Application No. PCT / US2019 / 031714, entitled “Digital Compensation for a Non-Linear System,” filed on May 10, 2019). Two example operating conditions that can be used to form an extended set of basis functions (i.e., in the case where d=2) can include a drain voltage (V) associated with at least one power amplifier of the wireless device. DD ), and temperature, in this case, the vector p[t] can be expressed as follows:

[0167] In some examples, Formation of a collateral set (comprising b·(d+1) basis functions, where b is the total number of terms in the initial / basic set of basis functions and d is the number of operating conditions used to generate the extended, collateral set of basis functions): a set of basis functions (containing b terms) is formed as the initial ("basic") set of unmodified basis functions, and d additional sets of basis functions (each having b basis function terms) are formed as the product of the corresponding measured operating conditions and the initial / basic set of basis functions. Thus, for a method involving the use of two (2) operating conditions (e.g., p1 and p2, corresponding to V DD and temperature) k (q u [t],p[t]) can be expressed as:

[0168] B 3(k-1)+1 (q) = B k0 (q)

[0169] B 3(k-1)+2 (q) = B k0 (q)·p1

[0170] B 3(k-1)+3( q)=B k0 (q)·p2

[0171] The extended set of basis functions can be generated in other ways, for example, by using other formulas and transformations to express the operating conditions (p1, p2, ..., p n ), where such formulas and transformations are applied to the initial / basis set of basis functions, instead of or in addition to the above formulas involving the product of the operating conditions of p1 and p2 and the basis set of basis functions. In addition, some extended basis function groups can be formed as functions of multiple operating conditions (e.g., each group of the extended set can be expanded to involve p1, p2, ..., pn Note that the expansion of the basis functions works best when there is some linear dependency between the operating conditions (eg, p1 and p2 in the above example).

[0172] In operation, a compensator such as compensator C 110 - based on a transmit chain (e.g., composed of Figure 1 The block F120 implements the inverse model operation of the nonlinear distortion of the DPD coefficients (implemented by the block F120 of FIG. 120 ) and uses an adaptive basis function weighted by regularly updated DPD coefficients (extended and adapted according to the values ​​of the operating conditions occurring simultaneously) - receives the input signal u and generates the intermediate input signal v as follows:

[0173]

[0174] Among them, f i (·) is the i-th basis function of the basis function, and x i is the i-th parameter (e.g., the i-th weight) corresponding to the i-th basis function. Each basis function (at least in the basis set) is a linear function of the input u (e.g., u(t-1)) or a nonlinear function (e.g., |u(t)| 2 ), nonlinear functions may include memory (e.g. u(t) * u(t-1)).

[0175] In some implementations, Figure 1 The adaptive process implemented by the system 100 adjusts the coefficients x of the predistorter according to the following by observing the measured samples y[t] and trying to minimize the regularized mean square error k (where k represents the use set x k The set of coefficients for a time instance or interval of :

[0176]

[0177] in,

[0178] In the above expression, e[t] = v[t] - y[t] is the input / output mismatch in the observed data, and ρ>0 is a regularization factor (used to prevent ill-conditioning of the optimization problem, which tends to lead to unstable interactions between the adaptive process and the application of the compensator and the optimal x * = argmin E(x)))), T = {t} is the set of time samples, and N is the number of elements in T.

[0179] When n is small (e.g., n<<100), minimization of E(x) can be performed by computing the following correlation matrix:

[0180]

[0181] where B[t] is a vector of basis functions (i.e., a vector of complex entries, one for each of the n basis functions):

[0182]

[0183] In some examples, the DPD coefficients weighting the basis functions can then be calculated according to the following expression, which includes a term that depends on p to regularize the optimization:

[0184]

[0185] Alternatively, in some embodiments (eg, when n is large), the DPD coefficients that weight the adaptive basis functions may be computed according to a stochastic gradient process. An example stochastic gradient process that may be used is as follows.

[0186] The function to be minimized E(x) can be expressed in the following form:

[0187]

[0188] Where a[t]=[B1(q y [t],p[t]),B2(q y [t],p[t]),...,B n (q y [t],p[t])] is a 1-by-n complex matrix formed by the basis function samples, and

[0189]

[0190] are the components of the Gramian matrix computed from the data.

[0191] As time goes by, the sample (q y [t],p[t],e[t]),t∈T is updated with the new measurement results. In some cases, this can be done by recalculating the Gramian, averaging the new and old Gramian, or replacing the old sample with the new one (q y [t],p[t],e[t]) samples to complete.

[0192] In some example embodiments, the adaptive process for DPD implementation attempts to find a vector x of compensator coefficients as an approximation of the optimal x*=argminE(x). The method of finding x*=argminE(x) is to calculate it as a linear equation (ρI n +G)x=L’s (unique) solution x*=(ρI n +G)- 1 L, where In This method requires the computation (and storage) of the n×n Hermitian matrix G, which is impractical even for very simple compensators (e.g., a compensator with n=100 complex coefficients would require the computation and storage of n copies of G). 2 = 10,000 real coefficients). Additionally, solving large systems of linear equations on an FPGA or ASIC is a challenge in itself and does not address the large storage challenges required by Gramians.

[0193] Alternatively, at point x∈C n The gradient of E = E(x) at is given by:

[0194]

[0195] and complex matrices that can be computed and stored as n×1, which can reduce storage requirements compared to those of Gramian.

[0196] Once calculated, the gradient can be used to Adjust the current guess at the optimal x makes it slightly better (as long as the step size parameter γ is small enough). Using the updated guess The gradient can be recalculated and another adjustment can be made. Repeating this process enough times leads to an arbitrarily accurate approximation of x*. Therefore, in methods based on the gradient of the error function E(x), the new / subsequent set of coefficients used to weight the basis functions applied by the compensator (e.g., compensator C 110) can be estimated / approximated as changes to the current coefficients by a predetermined multiple of the gradient value of E(x) (i.e., the step size γ).

[0197] Recalculating the gradient each time requires a large number of operations. Since the number of steps in the gradient search tends to be quite large before a satisfactory optimal level is reached, compensator adaptation based on a gradient search can be slow. Therefore, in some embodiments, in order to reduce the amount of computation required to derive the current value of the gradient of E(x) (based on which the updated coefficients are calculated), a sub-gradient method that minimizes E(x) can be used to approximate the DPD coefficients x. Specifically, based on the gradient The fact that is the expected value of a vector random variable implements the stochastic gradient method that minimizes the function E=E(x).

[0198] ξ=2[ρx+a[τ]'(a[τ]xe[τ])]

[0199] where τ is a random variable uniformly distributed over a finite set T = {t}, and a random sample of ξ in the gradient search iteration (which is easier to calculate than the true gradient in terms of computational effort). Much cheaper) can be replaced without a significant loss in convergence speed. In the above expressions, a[τ] represents the matrix of values ​​resulting from applying the basis functions weighted by the DPD coefficients x to the input samples at randomly selected instants corresponding to the random variable τ, and e[τ] is the input / output mismatch in the observed data between the output of the observer receiver and the output of the digital compensator (e.g., the output y of the sensing portion S130 and the output v of the compensator C 110).

[0200] Therefore, the stochastic gradient method that can be used involves constructing a sequence of random values (take the value in Cn), and at instance k (i.e., ) to update the current coefficient as follows:

[0201]

[0202] where τ1,τ2,...,τ n are independent random variables uniformly distributed over T = {t}, ρ>0 is a regularization constant according to the definition of E = E(x), and α>0 is a multiple (which can be a constant value that can be adjusted with regular or irregular periods) such that for each t∈T, α(ρ+|a[t]| 2 )<2. The value α indicates the value to be used for approximating the gradient A vector of random variables with the expected values ​​and the step size of the values.

[0203] When k→∞, the expected value can be shown to converge to x*=argminE(x). The average operation can be used according to the following expression:

[0204] where ∈∈(0,1]

[0205] For large k, as long as ∈>0 is small enough, and x * This method of minimizing E(x) is called a "projection" method because the mapping

[0206] x→x+|a[t]| -2 a[t]'(e[t]-a[t]x[t])

[0207] Project x onto the hyperplane a[t]x=e[t].

[0208] In a practical implementation of the process of computing an approximation of the gradient of E(x) and thus updating the current coefficients, τk is generated as a pseudo-random sequence of samples, and can be eliminated Calculation of (this formally corresponds to ∈ = 1, that is, ). Typically, this requires that for α(ρ+|a[t]| 2 ) uses a value of α with a smaller minimum upper bound (e.g., α(ρ+|a[t]| 2 )<1 or α(ρ+|a[t]| 2 )<0.5). In some embodiments, depending on the progress made by the stochastic gradient optimization process, the value of α is sometimes adjusted, where |e[τ k ]|and In some embodiments, periodic updating of the set of optimization problem parameters a[t], e[t] may be performed to replace data samples a[t], e[t] observed in the past with new observations.

[0209] Because the DPD coefficients are updated dynamically in the stochastic gradient method described herein (reflecting the fact that the behavior of the system 100 changes over time), the expected value for the random value ξ of the vector represents the gradient A good approximation or estimate of , thus allowing a cheaper calculation (from a computational effort perspective) of the value of the DPD coefficient x used by the compensator of system 100 .

[0210] In some embodiments, the value of the step size (multiplied by the step size for the gradient) may be adjusted based on a measure of the progress of the stochastic gradient process. determined approximation), where by comparing, for example, |e[τ k ]|and The measure of progress is determined by taking the average of the steps. In some embodiments, the process starts with a relatively large step size and then gradually decreases the step size.

[0211] Next about Figure 5 , shows a flow chart of an example process 500 for digital predistortion based on the use of adaptive basis functions (e.g., formed using operating conditions of a digital predistortion system performing process 500). Process 500 includes obtaining 510 a first set of digital predistortion (DPD) basis functions (shown as basic basis functions B k (q), which is expressed as a function of the samples of the input signal u, i.e., q u [t] = (u[t+l-1], u[t+l-2], ..., u[t+l-τ])), for controlling the operation of a digital predistorter of a wireless device operating on at least one received input signal. The at least one input signal is directed to a power amplification system (such as Figure 1 The system 100 includes a transmission chain (such as a power amplifier) ​​having at least one power amplifier that generates an output having nonlinear distortion. Figure 1 The sending part F 120).

[0212] Continue About Figure 5 , process 500 additionally includes determining an adaptation set of DPD basis functions derived from the first set of DPD basis functions, the adaptation set of DPD basis functions depending on operating conditions associated with operation of the wireless device (shown as provided by data p[t] received at block 520) for controlling operation of the digital predistorter. In some examples, the operating conditions may include one or more of the following, for example, channel information associated with a corresponding channel used by the wireless device to process at least one input signal, a corresponding temperature, a corresponding average power, a corresponding drain voltage (V DD ), a corresponding voltage standing wave ratio (VSWR) associated with the wireless device, and / or the age of the collected samples used to determine the DPD coefficient.

[0213] As discussed herein, an adaptive set of basis functions (if the adaptive set of basis functions is formed by transforming an original set of carefully selected basis functions into an extended set of basis functions, the original set comprising complex terms dependent on a τ-dimensional complex vector q (corresponding to complex input samples of the digital distorter)) is implemented, for example, by real terms dependent on a d-dimensional real vector p representing values ​​of operating conditions associated with operation of the wireless device on at least one received input signal (including environmental characteristics). The value d represents the number of parameters of operating conditions associated with operation of the wireless device for the basic set of extended basis functions. In such an embodiment, the total number of basis functions (in the extended set) used for digital predistortion of at least one input signal is proportional to the product of the number of terms b in the first set of basis functions and d+1, such that the total number of basis functions is proportional to b·(d+1). Thus, the basis functions include (d+1) groups of basis functions, wherein in some examples the first group of basis functions includes the first set of basis functions without any modification, and wherein each of the remaining d groups of basis functions is determined as a product of the first set of basis functions and one of the operating conditions (e.g., environmental characteristics) associated with the wireless device. In some other examples, other types of transformations expressed as linear or nonlinear functions of the operating conditions may be used to expand the base set of basis functions.

[0214] Continue About Figure 5 The process 500 also includes using the DPD coefficient (illustrated as the output x of block 530) k ) to configure a digital predistorter, wherein the DPD coefficients are based at least in part on an observed sample of the transmit chain in response to at least one input signal (e.g., Figure 1 and as described as input to block 530, and / or Figure 1The input signal u[t] or output v[t] of the compensator C 110 is determined for the adaptation set of DPD basis functions.

[0215] In some embodiments, configuring a predistorter having a DPD coefficient may include generating the DPD coefficient x according to k The DPD coefficients are calculated as weights of the basis functions according to an optimization process that minimizes an error function based on observation samples of the transmission chain and samples of at least one input signal. In such an embodiment, the DPD coefficients x are calculated according to the optimization process k This may include adaptively adjusting the DPD coefficient x according to the following minimization of the regularized mean square error function E(x): k :

[0216]

[0217] in, Where ρ>0 is the regularization factor, B k is the basis function for k=1 to n, q y [t] is the stack of input samples at time t, p[t] is a d-dimensional real vector of operating conditions (which may include environmental characteristics) at time t, v[t] is the output of the digital predistortion (DPD) compensator at time t, and y[t] is the output of the observation receiver coupled to the output of the transmit chain.

[0218] Adaptively adjust the DPD coefficient x k This may include deriving the correlation matrices G and L according to the following when n<<100, where n represents the number of basis functions:

[0219] in,

[0220] as well as

[0221] The DPD coefficient x is derived from k :

[0222]

[0223] Alternatively, the DPD coefficient x is adaptively adjusted k This may involve approximating the DPD coefficients x according to a stochastic gradient process k .

[0224] The third technique: Adaptive approach based on diversity

[0225] As discussed herein, another technique / approach to achieving agile estimation of predistortion parameters and / or resistance to rapidly changing operating conditions based on sparse data acquisition / availability is achieved by implementing a diversity matrix for maintaining a limited number of sample points based on which an adaptive process is performed. To facilitate robust digital predistortion performance, the diversity matrix technique described herein uses optimization data that includes input / output samples collected from a diverse set of operating conditions. The diversity (i.e., diversification) of the data used to perform the optimization process (e.g., using one or more optimization processes described herein) is achieved by using a diversity matrix that allows data samples that provide a greater diversity of data sets to be collected, and retains historical data samples that are highly relevant and can be used to update the DPD coefficients (the relevance of the samples may decrease over time as previously collected samples age).

[0226] Briefly, under the diversified matrix approach, a DPD system (such as Figure 1 The system 100 depicted in FIG. 1 collects data including data records (eg, maintained in a coupled adaptive portion A 140 or Figure 1 The data records may be stored in a memory storage device of any other subsystem of the transmit chain (e.g., implemented via transmit section F 120), wherein each data record includes a data record of input signal data and corresponding output signal data for a transmit chain (e.g., implemented via transmit section F 120). The records may be maintained in a matrix-like data structure, and each record may be associated with a set of operating condition values ​​(e.g., temperature, supply voltage associated with a power amplifier of the transmit section, load conditions, band attributes associated with a spectral band through which data is collected or transmitted, VSWR values, sampling rates, modulation modes, sample age, etc.), which may also be stored in the matrix. Each data record may be associated with a diverse measure / score representing the degree of distinguishability of the corresponding data record from other collected data records. Such measures may be stored as fields or entries of the data record (e.g., within a matrix storing other portions of the collected data), or may be stored as a separate data structure, wherein the locations of different measures within the data structure correspond to specific records associated with the different measures, respectively.

[0227] To manage the computational complexity of the optimization process involved in calculating the adaptive DPD coefficients (weighting basis functions such as the adaptive basis functions described herein), it may be necessary to select a subset of the collected samples. This selection may be necessary in the case where the optimization method is a method based on the use of a correlation matrix (as noted above, the use of a correlation matrix is computationally manageable when the number of data samples used in the optimization process is relatively small, e.g., n << 1000). Thus, in the diversity-based methods discussed herein, a subset of the records for optimization is selected based on a diversity measure associated with the data records. For example, in an embodiment where the diversity measure is associated with the degree of distinguishability / distinguishability between the records, the selection of the records to be used in the optimization may be a selection in which k records (from the available n records, where k < n) that provide the greatest distinguishability (i.e., the greatest diversity or diversification) are selected. In some implementations, the selection of the records is performed dynamically by determining whether to replace an existing record in the matrix structure associated with a particular diversity measure / score with a newly arrived record, where the diversity measure of the newly arrived record indicates a higher degree of distinguishability than the existing record upon receipt of the data. In this case, the previously stored record is replaced to support the new record. Further details regarding the process of replacing records (and thus maintaining records that increase or enhance the degree of record diversity) are provided below.

[0228] The calculation of the diversity measure (which may be performed by the adaptive portion A 140) may include, for each data record, calculating the diversity measure based on a distance metric representing the distance between the data elements representing the data for a particular signal instance (e.g., the values representing the samples and / or the values representing the operating conditions of the data record for the signal instance) and the corresponding data elements of the records for other signal instances (stored in the matrix). The distance may be calculated according to one or more formulas for determining distances in a mathematical hyperspace of size n. Other processes for calculating the diversity measure / score may also be used. The calculated diversity measures may then be ranked, and in implementations where only a certain number (k) of the available signal instances are used to derive the DPD coefficients, the k records having the highest diversity measure / score may be selected for use in the optimization process. Other criteria for selecting k records based on the diversity measure associated with the data records may be used.

[0229] As noted, the collected data records including the input and output samples and the corresponding operating conditions may be arranged in a matrix data structure, also referred to as the diversity database V, as follows:

[0230]

[0231] where q i = qy [t i ] is a complex vector of input samples collected for signal i (in some examples, 5, 10, or any number of samples may be collected for each signal instance i), p i = p[t i ] is the corresponding vector of operating conditions associated with the record corresponding to signal instance i, and e i =v[t i ]-y[t i ] is a compensator (such as, Figure 1 The difference (e) between the resulting predistortion output of the compensator C 110 and the observed output sample of the transmit chain i The values ​​are used in an optimization process to calculate adaptive DPD coefficients to weight the basis functions corresponding to the input signal u[t] provided to the input of the compensator). The diversity matrix V can be of any size, but in some typical examples, the number of columns m is less than 1000, which allows the optimization process for the adaptive process to be performed by, for example, the correlation matrix technique described herein. As noted, if the number of different sample records in the database (i different) is relatively large, the computational complexity for optimizing the DPD coefficients using the correlation matrix technique may be too high, and in such a case, a stochastic gradient process can be employed. The following V matrix example stored for each signal instance includes 3 samples and two operating condition parameters (e.g., VDD and temperature), resulting in a matrix defining a 5-dimensional space.

[0232]

[0233] However, typically each signal instance (stored in each column of the V matrix) will include more than three (3) signal sample points in order to improve the accuracy and robustness of the optimization process to derive the DPD coefficients x k .

[0234] In some embodiments, the matrix V may be accompanied by a minimum distance vector R, ie, R = [r1, r2, ..., r m], the minimum distance vector R maintains a diversity measure (e.g., one measure for each column / record of the diversity matrix V) that can be calculated as a distance function between the elements of a particular signal instance (e.g., a newly arrived signal instance) and any other column / record of the diversity matrix V. In such an embodiment, the greater the distance between any two records, the greater the diversity score (i.e., distinguishability) corresponding to the record used to optimize the DPD coefficient. In some embodiments, the minimum distance R vector entries can be combined with the V matrix (e.g., as added rows of the matrix V). As reflected by the vector R, large diversity measure values ​​can indicate a high degree of distinguishability between corresponding columns (which correspond to signal instances) and, therefore, a good distribution of samples over the space of possible operating conditions. Large diversity can reduce sampling bias that may be caused (or conversely increase the randomness of the samples) if the collected sample instances are too closely converged with each other (i.e., their corresponding operating conditions are too closely converged, which would result in a less robust predistorter).

[0235] As noted, in some embodiments, a matrix of predetermined size (e.g., having k columns and m rows, where each column represents a separate signal instance, and each row represents values ​​associated with the signal instance, including sample values ​​and values ​​representing operating conditions) is initially populated with an initial data set (e.g., all "0"), and, as new samples gradually arrive, their distances to the data entries stored in the V matrix are calculated (according to one or more distance formulas), and a determination is made as to whether the newly arrived samples provide more diversity (relative to the currently stored sample instances). If the newly arrived sample will result in a higher diversity of the data maintained in the V matrix, then the newly arrived data replaces one of the columns in the matrix V.

[0236] Determine (based on one or more criteria based on the computed diversity measure) whether to replace the data entry corresponding to the newly arrived data entry (q y An example of the process of replacing the existing data in the matrix V with data from a matrix [t], p[t], v[t]-y[t] is as follows:

[0237] 1) For i=1,…,m, calculate h i =dist * ((q i ,p i ),(q y [t],p[t])), where the approximate distance Represents the pairing (q,p) and In some embodiments, h i can be exported as q,p, The maximum absolute value of the difference between the corresponding individual components.

[0238] 2) Use min{r i ,h i}Replace each entry r of R i .

[0239] 3) Find the minimum entry r in the updated R k The index k∈{1,2,…,m} of R is replaced. In this operation, the replacement process identifies an entry from the updated R vector with the smallest minimum distance. This entry will correspond to a column whose relative distinguishability with respect to other sample data points is low (i.e., it is associated with a low diversity measure).

[0240] 4) According to q k =q y [t],p k =p[t],e k =v[t]-y[t], replace the kth column of the V matrix with the new column.

[0241] 5) Use h i Replace the kth entry r of R i .

[0242] As noted, in some embodiments, one example of an operating condition parameter that can be used is the age of the existing data samples. In such embodiments, relatively old existing sample data instances in the matrix V may result in large distance values ​​(i.e., large h values) between the old existing sample data instances and the new data sample points. Therefore, if a new data sample point replaces an old existing sample data instance in the matrix V, the corresponding entry in the R vector will have a relatively large h value, and therefore, the column that has just been replaced will be less likely to be replaced by the next sample data instance collected (i.e., because the corresponding entry is less likely to be the smallest entry in the R vector).

[0243] In some implementations, a decision may be made not to replace any existing columns with new sampled data instances. For example, if it is determined that the calculated diversity measure for the new sampled data instance will result in a less diverse set of sampled instances (e.g., because there is already a large set of sampled data points similar to the new sampled data point under consideration), the replacement process may determine to retain all existing records (e.g., matrix columns) and discard the newly collected sampled data instance.

[0244] Other methods and techniques of controlling (e.g., replacing) the contents of the V matrix (by applying different diversification criteria based on the calculated diversity measure) may be implemented. Such methods, including the above method of replacing columns of the V matrix, are configured to allow the most diverse data to be retained in the V matrix.

[0245] An adaptive process such as the various processes discussed herein may be performed at regular or irregular time intervals (e.g., every second, 10 seconds, 1 minute, 1 hour, etc.) using the available sampled data points (including new sampled data points that have been added to the V matrix since the previous update). Alternatively, in some examples, the adaptive process may be run after a certain number of updates have been performed on the V matrix (e.g., after 50% of the columns of the V matrix have been replaced since the previously run adaptive process). As discussed herein, an adaptive process for determining DPD coefficients for weighting basis functions may be performed by reoptimizing the DPD coefficients x according to the following:

[0246]

[0247] where m is the i-th column of the database matrix (i.e., V matrix), and x k (k=1, ..., n) are the elements of a complex n-dimensional vector of compensators (which act as a universal compensator when adapted using the method described herein), where

[0248] To illustrate the effect of using a diversity-based approach to select well-distributed sample data points to perform the DPD coefficient optimization process, consider Fig. 6A , Fig. 6A Graphs 610 and 612 are shown, which depict simulation results for an optimization process performed without ensuring a good diversity sample. In this simulation, a sample of 100 finite diverse input / output data with independent variable values ​​randomly scattered on [0,1] was generated. The distribution of the samples is Fig. 6A The optimization process of fitting a cubic polynomial is then performed. The fitting error (mismatch) caused by using limited diverse sample points is shown in Figure 612. As expected, the fitting process applied to limited diverse samples results in a large fitting error in the range of [-1, 1].

[0249] on the contrary, Figure 6B The simulation results are shown when the optimization process uses samples subjected to a diversity-based distribution method that promotes sample diversity. In this simulation, a diversified database matrix with 100 records (i.e., 100 columns) is continuously updated with new data. The first 1000 samples (with x and y independent variables) supplied and processed by the sample selection and replacement process (based on a method similar to that described above) have a range of x∈[-1,0], y∈[-1,0], then 1000 samples with x∈[-1,0], y∈[0,1] are processed, and finally 1000 samples with x∈[0,1], y∈[-1,0] are processed. As Figure 6BAs shown in Figure 620, the coverage process (i.e., the sample selection / replacement process) maintains moderately uniform coverage (and therefore high diversity / diversity levels) over the entire range [-1, 1]. Therefore, and as shown in Figure 622, relative to Fig. 6A The resulting fit mismatch of the simulation results illustrated in , is due to the optimization process performed (similar to that for Fig. 6A The fit mismatch caused by the optimization process applied in the simulation is greatly improved.

[0250] Next about Figure 7 , a flow chart of an example process 700 for digital predistortion (specifically, performing DPD adaptation using a diversity-based approach) is shown. The process 700 includes collecting 710 for including a transmission chain (e.g., similar to Figure 1 120) of a wireless device of the type depicted in the transmit chain, the transmit chain comprising at least one power amplifier producing an output having nonlinear distortion, wherein the data comprises each data record comprising input signal data for the transmit chain (e.g., signal u[t]) and corresponding output signal data (e.g., Figure 1 v[t] and y[t] shown in ) and also with the set of operating condition values ​​(shown as Figure 1 and 7 Each data record is associated with seven diverse measures representing the degree of distinguishability of the corresponding data record from other collected data records. Examples of operating conditions for each data record may include one or more of the following, for example, channel information associated with a corresponding channel used by the wireless device to process the corresponding data record, a corresponding temperature, a corresponding average power, a corresponding supply voltage (V DD ), a corresponding voltage standing wave ratio (VSWR) associated with the wireless device, and / or the age of each data record.

[0251] As noted, in some implementations, the predistorter may be implemented according to the first technical approach discussed above. In this approach, the predistorter may be adapted based on samples collected in a narrow subband rather than samples collected under wideband conditions (e.g., where the bandwidth of the wireless transceiver of the device acquiring the data is substantially equal to the spectral range for the predistorter to be implemented). A priori will be implemented for this purpose). Thus, in such an embodiment, collecting data for the wireless device may include collecting data in one or more spectral channels, the one or more spectral channels being within a wideband spectral range, wherein each of the one or more spectral channels has a corresponding bandwidth that is smaller than the range bandwidth of the spectral range.

[0252] In some examples, each data record may include one or more samples obtained by the wireless device under a particular set of operating conditions.Each data record may be maintained in a separate column of a record matrix data structure.

[0253] Continue About Figure 7 , process 700 also includes selecting 720 at least some of the data records according to a diversity criterion calculated based on a diversity measure for each of the data records. Selecting at least some of the data records according to a diversity criterion calculated based on a diversity measure for each of the data records may include selecting data records associated with respective diversity measures indicating high distinguishability in order to increase the diversity of the data records used to calculate the digital predistortion parameters.

[0254] The process 700 additionally includes calculating 730 digital predistortion parameters for controlling a digital predistorter of the wireless device based on the selected at least some of the data records, and applying 740 the digital predistorter configured with the calculated digital predistortion parameters to a subsequent input signal. In some embodiments, applying the digital predistorter configured with the digital predistortion parameters may include weighting basis functions of the digital predistorter, wherein the basis functions include (in embodiments such as those discussed with respect to adaptive extended basis functions formed from a basis set of basis functions) complex terms applied to values ​​of input samples provided in the data records, and wherein at least some of the basis functions also include real terms applied to values ​​of a set of operating conditions associated with the data records.

[0255] In some implementations, process 700 may also include, for each data record, calculating a diversity measure based on a distance metric representing a distance between a data element of the particular data record and a corresponding data element for at least some of the collected data records. The calculated diversity measure for each data record may be maintained in a distance vector data structure (such as the R vector discussed above) separate from a data structure storing the collected data records.

[0256] As discussed, under the diversification method, data samples used in the adaptive process are occasionally replaced with new sample data records that may be more relevant because they promote the diversity of samples used in the adaptive process (partially because they are updated). An example of a replacement process for data records includes occasionally collecting additional data records; in response to a determination that a diversity measure calculated for a particular one of the collected additional data records is more distinguishable than one of the previously stored data records relative to previously stored data records available at the record data structure, storing the particular one of the collected additional data records in the record data structure; and recalculating digital predistortion parameters at regular or irregular intervals based on the stored data samples available in the record data structure, the stored data samples available in the record data structure including the stored particular one of the collected additional data records. In such an embodiment, storing the particular one of the collected additional data records may include replacing one of the previously stored data records in the record data structure with the particular one of the collected additional data records. The process may also include calculating a corresponding diversity measure for a particular one of the collected additional data records, including determining, for each pair of records including the particular one of the collected additional data records and at least one previously stored data record, a maximum absolute value of differences between corresponding elements of the data records.

[0257] The method described above can be used in conjunction with the techniques described in PCT Application No. PCT / US2019 / 031714, entitled “Digital Compensation for a Non-Linear System,” filed on May 10, 2019, which is incorporated herein by reference. For example, the techniques described in that application can be used to implement an actuator (referred to as a predistorter in the incorporated application) and adapt its parameters, and specifically form the actuator to respond to an envelope signal or other signal related to power control of a power amplifier.

[0258] In some implementations, a computer-accessible non-transitory storage medium includes a database (also referred to as a "design structure" or "integrated circuit definition data set") that represents a system that includes some, some, or all of the components of the digital predistortion and / or DPD adaptation implementation and process described herein. In general, a computer-accessible storage medium may include any non-transitory storage medium that can be accessed by a computer during use to provide instructions and / or data to the computer. For example, a computer-accessible storage medium may include storage media such as magnetic or optical disks and semiconductor memories. In general, a database representing a system may be a database or other data structure that can be read by a program and used directly or indirectly to manufacture hardware including the system. For example, the database may be a behavioral level description or a register transfer level (RTL) description of hardware functions in a high-level design language (HDL) such as Verilog or VHDL. The description may be read by a synthesis tool that may synthesize the description to generate a netlist including a list of gates from a synthesis library. The netlist includes a set of gates that also represent the functions of the hardware including the system. The netlist may then be placed and routed to generate a data set for describing the geometry to be applied to the mask. The masks may then be used in various semiconductor fabrication steps to produce one or more semiconductor circuits corresponding to the system.In other examples, the database itself may be a netlist (with or without a synthesis library) or a data set.

[0259] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as the meaning commonly or conventionally understood. As used herein, the articles "a" and "an" refer to one or more than one (i.e., at least one) of the grammatical object of the article. By way of example, "element" refers to an element or more than one element. When referring to measurable values ​​such as amount and duration, "about" and / or "approximately" as used herein encompass changes relative to ±20% or ±10%, ±5% or ±0.1% of a specified value, because such changes are appropriate in the context of systems, devices, circuits, methods, and other implementations described herein. When referring to measurable values ​​such as amount, duration, and physical properties (such as frequency, etc.), "substantially" as used herein also encompasses changes relative to ±20% or ±10%, ±5% or ±0.1% of a specified value, because such changes are appropriate in the context of systems, devices, circuits, methods, and other implementations described herein.

[0260] As used herein (including in the claims), "or" used in a list of items preceded by "at least one of" or "one or more of" indicates a disjunctive list, so that, for example, a list of "at least one of A, B, and C" means A or B or C or AB or AC or BC or ABC (i.e., A and B and C), or a combination of more than one feature (e.g., AA, AAB, ABBC, etc.). Furthermore, as used herein, unless otherwise stated, a statement that a function or operation is "based on" a term or condition means that the function or operation is based on the stated term or condition, and may be based on one or more terms and / or conditions in addition to the stated term or condition.

[0261] Although specific embodiments are disclosed in detail herein, this is done by way of example only for the purpose of illustration and is not intended to limit the scope of the invention, which is defined by the scope of the appended claims. The features of the disclosed embodiments may be combined, rearranged, etc. within the scope of the invention to produce more embodiments. Some other aspects, advantages, and modifications are considered to be within the scope of the claims provided herein. The claims presented represent at least some of the embodiments and features disclosed herein. Other unrequired embodiments and features are also contemplated.

Claims

1. A method for digital predistortion, include: collecting data for a wireless device including a transmit chain including at least one power amplifier producing an output having nonlinear distortion, the data comprising data records, wherein each data record includes input signal data and corresponding output signal data for the transmit chain and is also associated with a set of operating condition values, wherein each data record is associated with a diversity measure representing a degree of distinguishability of the corresponding data record from other collected data records; For each of the data records, calculating the diversity measure based on a distance metric, wherein the distance metric represents a distance between a data element of a particular data record and corresponding data elements of at least some of the collected data records; selecting at least some of the data records according to a diversity criterion calculated based on the diversity measure for the respective data records; calculating digital predistortion parameters for controlling a digital predistorter of the wireless device based on selected at least some of the data records; and The digital predistorter configured with the calculated digital predistortion parameters is applied to a subsequent input signal.

2. The method according to claim 1, in, The set of operating conditions for each data record includes one or more of the following: channel information associated with a corresponding channel used by the wireless device to process the corresponding data record, a corresponding temperature, a corresponding average power, a corresponding supply voltage (V) associated with at least one power amplifier of the transmission chain, and a corresponding power supply voltage (V) associated with at least one power amplifier of the transmission chain. DD , a corresponding voltage standing wave ratio (VSWR) associated with the wireless device, and the age of each data record.

3. The method according to claim 1, in, The data collected from the wireless device includes: The data is collected in one or more spectral channels, the one or more spectral channels being located within a wideband spectral range, wherein each of the one or more spectral channels has a respective bandwidth that is less than a range bandwidth of the spectral range.

4. The method according to claim 1, in, Each of the data records includes one or more samples obtained by the wireless device under a particular set of operating conditions.

5. The method according to claim 4, in, Each of the data records is maintained in a separate column of a record matrix data structure.

6. The method according to claim 1, in, Selecting at least some of the data records according to the diversity criterion calculated based on the diversity measure of the data records comprises: Data records associated with respective diversity measures indicative of high distinguishability are selected to increase the diversity of data records used to calculate the digital predistortion parameters.

7. The method according to claim 1, further comprising: include: The calculated diversity measure for each of the data records is maintained in a distance vector data structure that is separate from a data structure used to store the collected data records.

8. The method according to claim 1, in, The digital predistorter configured with the digital predistortion parameters comprises: Basis functions of the digital predistorter are weighted, wherein the basis functions include complex terms applied to values ​​of input samples provided in the data record, and wherein at least some of the basis functions also include real terms applied to values ​​of a set of operating conditions associated with the data record.

9. A method for digital predistortion, include: collecting data for a wireless device including a transmit chain including at least one power amplifier producing an output having nonlinear distortion, the data comprising data records, wherein each data record includes input signal data and corresponding output signal data for the transmit chain and is also associated with a set of operating condition values, wherein each data record is associated with a diversity measure representing a degree of distinguishability of the corresponding data record from other collected data records; selecting at least some of the data records according to a diversity criterion calculated based on the diversity measure for the respective data records; calculating digital predistortion parameters for controlling a digital predistorter of the wireless device based on selected at least some of the data records; applying the digital predistorter configured with the calculated digital predistortion parameters to a subsequent input signal; Collect additional data records; In response to determining that a diversity measure calculated for a particular one of the collected additional data records is more distinguishable than data records in the previously stored data records relative to previously stored data records available at a record data structure, storing the particular one of the collected additional data records in the record data structure; and The digital predistortion parameters are recalculated at regular or irregular intervals based on stored data samples available in the record data structure, wherein the stored data samples include the stored specific additional data record among the collected additional data records.

10. The method according to claim 9, in, Storing the specific additional data record among the collected additional data records comprises: A data record in the previously stored data records is replaced in the record data structure with the particular additional data record in the collected additional data records.

11. The method according to claim 9, further comprising: include: Calculating a corresponding diversity measure for the particular one of the collected additional data records includes determining, for each pair of records including the particular one of the collected additional data records and at least one of the previously stored data records, a maximum absolute value of differences between corresponding elements of the data records.

12. The method according to claim 9, in, The set of operating conditions for each data record includes one or more of the following: channel information associated with a corresponding channel used by the wireless device to process the corresponding data record, a corresponding temperature, a corresponding average power, a corresponding supply voltage (V) associated with at least one power amplifier of the transmission chain, and a corresponding power supply voltage (V) associated with at least one power amplifier of the transmission chain. DD , a corresponding voltage standing wave ratio (VSWR) associated with the wireless device, and the age of each data record.

13. The method according to claim 9, in, The data collected from the wireless device includes: The data is collected in one or more spectral channels, the one or more spectral channels being located within a wideband spectral range, wherein each of the one or more spectral channels has a respective bandwidth that is less than a range bandwidth of the spectral range.

14. The method according to claim 9, in, Each of the data records includes one or more samples obtained by the wireless device under a particular set of operating conditions.

15. The method according to claim 14, in, Each of the data records is maintained in a separate column of a record matrix data structure.

16. The method according to claim 9, in, Selecting at least some of the data records according to the diversity criterion calculated based on the diversity measure of the data records comprises: Data records associated with respective diversity measures indicative of high distinguishability are selected to increase the diversity of data records used to calculate the digital predistortion parameters.

17. The method according to claim 9, in, The digital predistorter configured with the digital predistortion parameters includes: Basis functions of the digital predistorter are weighted, wherein the basis functions include complex terms applied to values ​​of input samples provided in the data record, and wherein at least some of the basis functions also include real terms applied to values ​​of a set of operating conditions associated with the data record.

18. A method for digital predistortion, include: collecting data for a wireless device including a transmit chain including at least one power amplifier producing an output having nonlinear distortion, the data comprising data records, wherein each data record includes input signal data and corresponding output signal data for the transmit chain and is also associated with a set of operating condition values, wherein each data record is associated with a diversity measure representing a degree of distinguishability of the corresponding data record from other collected data records; selecting at least some of the data records according to a diversity criterion calculated based on the diversity measure for the respective data records; calculating digital predistortion parameters for controlling a digital predistorter of the wireless device based on selected at least some of the data records; applying the digital predistorter configured with the calculated digital predistortion parameters to a subsequent input signal, Wherein, applying the digital predistorter configured with the digital predistortion parameters comprises: deriving basis functions of the digital predistorter from a first basis function set, the basis functions of the digital predistorter being a basis function extension set comprising (d+1) groups of basis functions, wherein d represents the number of parameters of the operating condition, wherein a first group of the basis function extension set comprises the first basis function set without any modification, and wherein each of the remaining d groups of the basis function extension set is determined as a product of the first basis function set and one of the parameters of the operating condition, and Basis functions of the digital predistorter are weighted, wherein the basis functions include complex terms applied to values ​​of input samples provided in the data record, and wherein at least some of the basis functions also include real terms applied to values ​​of a set of operating conditions associated with the data record.

19. The method according to claim 18, in, The set of operating conditions for each data record includes one or more of the following: channel information associated with a corresponding channel used by the wireless device to process the corresponding data record, a corresponding temperature, a corresponding average power, a corresponding supply voltage (V) associated with at least one power amplifier of the transmission chain, and a corresponding power supply voltage (V) associated with at least one power amplifier of the transmission chain. DD , a corresponding voltage standing wave ratio (VSWR) associated with the wireless device, and the age of each data record.

20. The method according to claim 18, in, The data collected from the wireless device includes: The data is collected in one or more spectral channels, the one or more spectral channels being located within a wideband spectral range, wherein each of the one or more spectral channels has a respective bandwidth that is less than a range bandwidth of the spectral range.

21. The method according to claim 18, in, Each of the data records includes one or more samples obtained by the wireless device under a particular set of operating conditions.

22. The method according to claim 21, in, Each of the data records is maintained in a separate column of a record matrix data structure.

23. The method according to claim 18, in, Selecting at least some of the data records according to the diversity criterion calculated based on the diversity measure of the data records comprises: Data records associated with respective diversity measures indicative of high distinguishability are selected to increase the diversity of data records used to calculate the digital predistortion parameters.

24. A linearization system for performing digital predistortion, the system include: a transmit chain comprising at least one power amplifier producing an output having non-linear distortion; as well as A digital predistorter comprising an adaptive part and a compensator, the digital predistorter being configured to: collecting data comprising data records, wherein each data record comprises input signal data and corresponding output signal data for the transmission chain and is also associated with a set of operating condition values, wherein each data record is associated with a diversity measure representing a degree of distinguishability of the corresponding data record from other collected data records, For each of the data records, calculating the diversity measure based on a distance metric, wherein the distance metric represents a distance between a data element of a particular data record and corresponding data elements of at least some of the collected data records, selecting at least some of the data records according to a diversity criterion calculated based on the diversity measure of the data records, calculating digital predistortion parameters for controlling the digital predistorter based on selected at least some of the data records, and The subsequent input signal is digitally predistorted according to the calculated digital predistortion parameters.

25. The system according to claim 24, in, The digital predistorter configured to select at least some of the data records according to the diversity criterion calculated based on the diversity measure of the data records is further configured to: Data records associated with respective diversity measures indicative of high distinguishability are selected to increase the diversity of data records used to calculate the digital predistortion parameters.

26. The system of claim 24, wherein the set of operating conditions for each data record comprises one or more of the following: channel information associated with a corresponding channel used by the wireless device to process a corresponding data record, a corresponding temperature, a corresponding average power, a corresponding supply voltage (V) associated with at least one power amplifier of the transmission chain, and DD , a corresponding voltage standing wave ratio (VSWR) associated with the wireless device, and the age of each data record.

27. A linearization system for performing digital predistortion, the system include: a transmit chain comprising at least one power amplifier producing an output having non-linear distortion; as well as A digital predistorter, comprising an adaptive part and a compensator, wherein the digital predistorter is configured to: collecting data comprising data records, wherein each data record comprises input signal data and corresponding output signal data for the transmission chain and is also associated with a set of operating condition values, wherein each data record is associated with a diversity measure representing a degree of distinguishability of the corresponding data record from other collected data records, selecting at least some of the data records according to a diversity criterion calculated based on the diversity measure of the data records, calculating digital predistortion parameters for controlling the digital predistorter based on selected at least some of the data records, The subsequent input signal is digitally predistorted according to the calculated digital predistortion parameters. Collect additional data records, In response to determining that the diversity measure calculated for a particular one of the collected additional data records is more distinguishable than data records in the previously stored data records relative to previously stored data records available at the record data structure, storing the particular one of the collected additional data records in the record data structure, and The digital predistortion parameters are recalculated at regular or irregular intervals based on stored data samples available in the record data structure, wherein the stored data samples include the stored specific additional data record among the collected additional data records.

28. The system according to claim 27, in, The digital predistorter configured to select at least some of the data records according to the diversity criterion calculated based on the diversity measure of the data records is further configured to: Data records associated with respective diversity measures indicative of high distinguishability are selected to increase the diversity of data records used to calculate the digital predistortion parameters.

29. The system of claim 27, wherein the set of operating conditions for each data record comprises one or more of the following: channel information associated with a corresponding channel used by the wireless device to process a corresponding data record, a corresponding temperature, a corresponding average power, a corresponding supply voltage (V) associated with at least one power amplifier of the transmission chain, and DD , a corresponding voltage standing wave ratio (VSWR) associated with the wireless device, and the age of each data record.

30. A linearization system for performing digital predistortion, the system include: a transmit chain comprising at least one power amplifier producing an output having non-linear distortion; as well as A digital predistorter, comprising an adaptive part and a compensator, wherein the digital predistorter is configured to: collecting data comprising data records, wherein each data record comprises input signal data and corresponding output signal data for the transmission chain and is also associated with a set of operating condition values, wherein each data record is associated with a diversity measure representing a degree of distinguishability of the corresponding data record from other collected data records, selecting at least some of the data records according to a diversity criterion calculated based on the diversity measure of the data records, calculating digital predistortion parameters for controlling the digital predistorter based on selected at least some of the data records, Performing digital predistortion on a subsequent input signal according to the calculated digital predistortion parameter, wherein the digital predistorter configured to perform digital predistortion on the subsequent input signal is further configured to: deriving an extended set of basis functions of the digital predistorter from a first set of basis functions, the extended set of basis functions comprising (d+1) groups of basis functions, wherein d represents the number of parameters of the operating conditions, wherein a first group of the extended set of basis functions comprises the first set of basis functions without any modification, and wherein each of the remaining d groups of the extended set of basis functions is determined as a transformation of the first set of basis functions according to a function of at least one of the parameters of the operating conditions, and The calculated digital predistortion parameters are used to weight the basis function extension set, wherein the basis functions include complex terms applied to values ​​of input samples provided in the data record, and wherein at least some of the basis functions also include real terms applied to values ​​of a set of operating conditions associated with the data record.

31. The system according to claim 30, in, The digital predistorter configured to select at least some of the data records according to the diversity criterion calculated based on the diversity measure of the data records is further configured to: Data records associated with respective diversity measures indicative of high distinguishability are selected to increase the diversity of data records used to calculate the digital predistortion parameters.

32. The system of claim 30, wherein the set of operating conditions for each data record comprises one or more of the following: channel information associated with a corresponding channel used by the wireless device to process a corresponding data record, a corresponding temperature, a corresponding average power, a corresponding supply voltage (V) associated with at least one power amplifier of the transmission chain, and DD , a corresponding voltage standing wave ratio (VSWR) associated with the wireless device, and the age of each data record.

33. A non-transitory machine-readable medium storing a program for causing a computer to execute the method according to any one of claims 1 to 23.