阅读说明：本技术 每分支的、组合的和分组组合的mimo dpd (Per-branch, combined and packet-combined MIMO DPD ) 是由 M·哈米德 于 2019-03-15 设计创作，主要内容包括：本文公开了用于多输入多输出(MIMO)发射机中的数字预失真(DPD)的系统和方法。在一些实施例中,MIMO发射机包括多个天线分支,多个天线分支包括耦合到相应多个天线元件的相应多个功率放大器。MIMO发射机还包括一个或多个DPD系统,该一个或多个DPD系统可操作以对相应的一组或多组输入信号进行预失真,以为相应的一组或多组天线分支提供相应的一组或多组预失真输入信号。每组天线分支包括多个天线分支中的至少两个。在一些实施例中,MIMO发射机是大规模MIMO发射机。还公开了用于使用迭代学习控制(ILC)和内核回归的MIMO发射机的每分支DPD方案的实施例。(Systems and methods for Digital Predistortion (DPD) in a multiple-input multiple-output (MIMO) transmitter are disclosed herein. In some embodiments, a MIMO transmitter includes a plurality of antenna branches including a respective plurality of power amplifiers coupled to a respective plurality of antenna elements. The MIMO transmitter further includes one or more DPD systems operable to predistort the respective one or more sets of input signals to provide the respective one or more sets of predistorted input signals for the respective one or more sets of antenna branches. Each group of antenna branches includes at least two of the plurality of antenna branches. In some embodiments, the MIMO transmitter is a massive MIMO transmitter. Embodiments of a DPD per branch scheme for MIMO transmitters using Iterative Learning Control (ILC) and kernel regression are also disclosed.)

1. A multiple-input multiple-output, MIMO, transmitter (500, 1200), comprising:

a plurality of antenna branches (502, 1202) including a respective plurality of power amplifiers (506, 1202) coupled to a respective plurality of antenna elements (508, 1204); and

one or more digital predistortion DPD systems (510, 1206) operable to, for each antenna branch of the one or more groups of antenna branches:

determining one or more model parameters for a combined MIMO DPD scheme for the set of antenna branches, the combined MIMO DPD scheme being an iterative learning control ILC combined MIMO DPD scheme with kernel regression; and

predistorting a set of input signals for the set of antenna branches based on the determined one or more model parameters in accordance with the combined MIMO DPD scheme for the set of antenna branches to provide a corresponding set of predistorted input signals for the set of antenna branches;

wherein each group of antenna branches comprises at least two of the plurality of antenna branches (502, 1202).

2. The MIMO transmitter (500, 1200) of claim 1, wherein:

the one or more sets of antenna branches comprise two or more sets of antenna branches that are disjoint subsets of the plurality of antenna branches (502, 1202); and

the one or more DPD systems (510, 1206) comprise two or more DPD systems (1206-1 to 1206- (L/S)) operable to predistort respective two or more sets of input signals to provide respective two or more sets of predistorted input signals for the two or more sets of antenna branches, respectively.

3. The MIMO transmitter (500, 1200) of claim 2, wherein each DPD system (1206) of the two or more DPD systems (1206-1 to 1206- (L/S)) is further operable to:

obtaining (1400) N samples of each of the plurality of input signals in a respective one of the respective two or more sets of input signals;

determining (1402-1410) a desired combined input signal U for a respective one of the two or more sets of antenna branches_s；

Generating (1412) a kernel regression matrix, θ, for the respective one of the two or more sets of antenna branches based on the N samples of each of the plurality of input signals in the respective one of the two or more sets of input signals_s(ii) a And

based on the kernel regression matrix θ for the respective one of the two or more sets of antenna branches_sAnd said desired combined input signal U_sCalculating (1414) one or more model parameters of a DPD model used by the DPD system (1206) to predistort the plurality of input signals in the respective one of the respective two or more sets of input signals.

4. Root of herbaceous plantThe MIMO transmitter (500, 1200) of claim 3, wherein to determine (1402) the desired combined input signal U for the respective one of the two or more sets of antenna branches (1410)_sEach DPD system (1206) of the two or more DPD systems (1206-1 to 1206- (L/S)) is further operable to:

initializing (1402) the desired combined input signal U for the respective one of the two or more sets of antenna branches_s；

Iteratively performing the following operations until at least one predefined criterion is met:

determining (1406) an error E based on a difference between_s: (a) a desired combined output signal of the respective one of the two or more sets of antenna branches and (b) when the desired combined input signal U is_sAn actual combined output signal of the respective one of the two or more sets of antenna branches when applied to the respective one of the two or more sets of antenna branches; and

based on the error E_sUpdating (1408) the desired combined input signal U for the respective one of the two or more sets of antenna branches_s。

5. The MIMO transmitter (500, 1200) of claim 3 or 4, wherein the one or more model parameters are weights applied by the DPD system (1206) to predistort the plurality of input signals in the respective one of the respective two or more sets of input signals.

6. The MIMO transmitter (500, 1200) of claim 1, wherein:

the one or more groups of antenna branches consist of a single group of antenna branches comprising the plurality of antenna branches (502); and

the one or more DPD systems (510, 1206) consist of a single DPD system (510), the single DPD system (510) operable to predistort a plurality of input signals using an ILC combined MIMO DPD scheme for the plurality of antenna branches (502) with kernel regression to provide a plurality of predistorted input signals for the plurality of antenna branches (502).

7. The MIMO transmitter (500, 1200) of claim 6, wherein the single DPD system (510) is further operable to:

obtaining (700) N samples of each of the plurality of input signals;

determining (702-710) a desired combined input signal U for the plurality of antenna branches (502);

generating (712) a kernel regression matrix θ for the plurality of antenna branches (502) based on the N samples of each of the plurality of input signals; and

calculating (714) one or more model parameters for a DPD model utilized by the single DPD system (510) based on the kernel regression matrix θ and the desired combined input signal U to predistort the plurality of input signals.

8. The MIMO transmitter (500, 1200) of claim 7, wherein to determine (702 and 710) the desired combined input signal U for the plurality of antenna branches (502), the single DPD system (510) is further operable to:

initializing (702) the desired combined input signal U for the plurality of antenna branches (502);

iteratively performing the following operations until at least one predefined criterion is met:

determining (706) an error E based on a difference between: (a) a desired combined output signal of the plurality of antenna branches (502) and (b) an actual combined output signal of the plurality of antenna branches (502) when the desired combined input signal U is applied to the plurality of antenna branches (502); and

updating (708) the desired combined input signal U for the plurality of antenna branches (502) based on the error E.

9. The MIMO transmitter (500, 1200) of claim 7 or 8, wherein the one or more model parameters are weights applied by the single DPD system (510) to predistort the plurality of input signals.

10. The MIMO transmitter (500, 1200) of any of claims 1 to 9, wherein the ILC combined MIMO DPD scheme with kernel regression or the ILC combined MIMO DPD scheme with kernel regression for a respective one of the respective two or more sets of antenna branches uses a radial basis function, RBF, kernel.

11. The MIMO transmitter (500, 1200) of any of claims 1 to 10, wherein the ILC combined MIMO DPD scheme with kernel regression or the ILC combined MIMO DPD scheme with kernel regression for the respective one of the respective two or more sets of antenna branches uses a generalized memory polynomial GMP base as a kernel.

12. The MIMO transmitter (500, 1200) of any of claims 1 to 11, wherein the ILC combined MIMO DPD scheme with kernel regression or the ILC combined MIMO DPD scheme with kernel regression for the respective one of the respective two or more sets of antenna branches, with kernel regression, takes into account memory effects.

13. A method of performing digital predistortion DPD in a multiple-input multiple-output, MIMO, transmitter (500, 1200), the transmitter (500, 1200) comprising a plurality of antenna branches (502, 1202), the plurality of antenna branches (502, 1202) comprising a respective plurality of power amplifiers (506, 1202) coupled to a respective plurality of antenna elements (508, 1204), the method comprising:

for each of the one or more sets of antenna branches:

determining (600, 1300) one or more model parameters for a combined MIMO DPD scheme for the set of antenna branches, the combined MIMO DPD scheme being an iterative learning control ILC combined MIMO DPD scheme with kernel regression; and

predistorting (602, 1302) a set of input signals for the set of antenna branches based on the determined one or more model parameters in accordance with the combined MIMO DPD scheme for the set of antenna branches to provide a corresponding set of predistorted input signals for the set of antenna branches;

wherein each group of antenna branches comprises at least two of the plurality of antenna branches (502, 1202).

14. The method of claim 13, wherein the one or more sets of antenna branches comprise two or more sets of antenna branches, the two or more sets of antenna branches being disjoint subsets of the plurality of antenna branches (502, 1202).

15. The method of claim 14, wherein determining (1300), for each of the two or more sets of antenna branches, the one or more model parameters of the combined MIMO DPD scheme comprises:

obtaining (1400) N samples for each of a plurality of input signals in the set of input signals;

determining (1402-1410) a desired combined input signal U for the set of antenna branches_s；

Generating (1412) a kernel regression matrix θ for the set of antenna branches based on the N samples of each of the plurality of input signals in the set of input signals_s(ii) a And

based on the kernel regression matrix θ for the set of antenna branches_sAnd said desired combined input signal U_s-calculating (1414) the one or more model parameters for the combined MIMO DPD scheme for the set of antenna branches.

16. The method of claim 15, wherein for the two or more sets of daysDetermining (1402-1410) for each group of antenna branches the desired combined input signal U for that group of antenna branches_sThe method comprises the following steps:

initializing (1402) the desired combined input signal U for the respective one of the two or more sets of antenna branches_s；

Iteratively performing the following operations until at least one predefined criterion is met:

determining (1406) an error E based on a difference between_s: (a) a desired combined output signal of the set of antenna branches and (b) a desired combined input signal U_sAn actual combined output signal of the set of antenna branches when applied to the respective one of the two or more sets of antenna branches; and

based on the error E_sUpdating (1408) the desired combined input signal U for the set of antenna branches_s。

17. The method of claim 15 or 16, wherein the one or more model parameters are weights applied to predistort the plurality of input signals in the set of input signals.

18. The method of claim 13, wherein:

the one or more sets of antenna branches consists of a single set of antenna branches, the single set of antenna branches including the plurality of antenna branches (502, 1202); and

the combined MIMO DPD scheme for the set of antenna branches is an iterative learning control, ILC, combined MIMO DPD scheme for the plurality of antenna branches (502, 1202) with kernel regression.

19. The method of claim 18, wherein determining (600) the one or more model parameters of the combined MIMO DPD scheme for the single set of antenna branches comprises:

obtaining (700) N samples for each of a plurality of input signals for the single set of antenna branches;

determining (702-710) a desired combined input signal U for the plurality of antenna branches (502) in the single set of antenna branches;

generating (712) a kernel regression matrix θ for the plurality of antenna branches (502) based on the N samples of each of the plurality of input signals; and

calculating (714) the one or more model parameters for the ILC combined MIMO DPD scheme based on the kernel regression matrix θ and the desired combined input signal U.

20. The method as recited in claim 19, wherein determining (702-710) said desired combined input signal U for said plurality (502) of antenna branches in said single set of antenna branches comprises, for said single set of antenna branches:

initializing (702) the desired combined input signal U for the plurality of antenna branches (502);

iteratively performing the following operations until at least one predefined criterion is met:

determining (706) an error E based on a difference between: (a) a desired combined output signal of the plurality of antenna branches (502) and (b) an actual combined output signal of the plurality of antenna branches (502) when the desired combined input signal U is applied to the plurality of antenna branches (502); and

updating (708) the desired combined input signal U for the plurality of antenna branches (502) based on the error E.

21. The method of claim 19 or 20, wherein the one or more model parameters for the ILC combined MIMO DPD scheme are weights applied to predistort the plurality of input signals.

22. The method of any of claims 13 to 21, wherein the ILC or the ILC combined MIMO DPD scheme with kernel regression for a respective one of two or more respective sets of antenna branches uses a radial basis function RBF kernel.

23. The method of any of claims 13 to 22, wherein the ILC combined MIMO DPD scheme with kernel regression or the ILC combined MIMO DPD scheme with kernel regression for the respective one of the respective two or more sets of antenna branches uses a generalized memory polynomial GMP base as kernel.

24. The method of any of claims 13-23, wherein the ILC combined MIMO DPD scheme with kernel regression or the ILC combined MIMO DPD scheme with kernel regression for the respective one of the respective two or more sets of antenna branches accounts for memory effects.

25. A multiple-input multiple-output, MIMO, transmitter (100), comprising:

a plurality of antenna branches (102) comprising a respective plurality of power amplifiers (106) coupled to a respective plurality of antenna elements (108); and

a plurality of iterative learning control ILC digital predistortion DPD systems (110) operable to predistort a plurality of input signals for the plurality of antenna branches (102), respectively, using an ILC per branch MIMO DPD scheme with kernel regression.

26. The MIMO transmitter (100) of claim 25, wherein each ILC DPD system (110) of the plurality of ILC DPD systems (110) is further operable to:

obtaining (300) N samples of a respective one of the plurality of input signals;

determining (302) an expected input signal U for a respective one of the plurality of antenna branches (102)_l；

Based on the N samples of the respective one of the plurality of input signalsGenerating (312) a kernel regression matrix θ for the respective one of the plurality of antenna branches (102)_l(ii) a And

based on the kernel regression matrix θ for the respective one of the plurality of antenna branches (102)_lAnd said desired input signal U_lAnd calculating (314) one or more model parameters for a DPD model used by the ILC DPD system (110) to predistort the respective one of the plurality of input signals.

27. The MIMO transmitter (100) of claim 26, wherein each ILC DPD system (110) of the plurality of ILC DPD systems (110) is further operable to generate (312) the kernel regression matrix θ for the respective one of the plurality of antenna branches (102) based on (a) the N samples of the respective one of the plurality of input signals and (b) N samples from one or more additional input signals of the plurality of input signals_l。

28. The MIMO transmitter (100) of claim 26 or 27, wherein for determining (302) the desired input signal U for a respective one of the plurality of antenna branches (102)_lEach ILC DPD system (110) of the plurality of ILC DPD systems (110) further operable to:

initializing (302) the desired input signal U of the respective one of the plurality of antenna branches (102)_l；

Iteratively performing the following operations until at least one predefined criterion is met:

determining (306) an error E based on a difference between_l: (a) a desired output signal of said respective one of said plurality of antenna branches (102) and (b) a desired input signal U_lAn actual output signal of the respective one of the plurality of antenna branches (102) when applied to the respective one of the plurality of antenna branches (102)；

Based on the error E_lUpdating (308) the desired input signal U for the respective one of the plurality of antenna branches (102)_l。

29. The MIMO transmitter (100) of claim 26 or 28, wherein the one or more model parameters are weights applied by the ILC DPD system (110) to predistort the respective one of the plurality of input signals.

30. The MIMO transmitter (100) of any of claims 25 to 29, wherein the ILC per branch MIMO DPD scheme with kernel regression uses radial basis functions RBF kernels.

31. The MIMO transmitter (100) of any of claims 25 to 30, wherein the ILC per branch MIMO DPD scheme with kernel regression takes into account memory effects on a predefined memory depth.

32. The MIMO transmitter (100) of any of claims 25 to 31, wherein the ILC per branch MIMO DPD scheme takes into account antenna coupling effects.

33. A method of performing digital predistortion, DPD, in a multiple-input multiple-output, MIMO, transmitter (100), the transmitter (100) comprising a plurality of antenna branches (102), the plurality of antenna branches (102) comprising a respective plurality of power amplifiers (106) coupled to a respective plurality of antenna elements (108), the method comprising:

determining (200) one or more model parameters for each antenna branch (102) of the plurality of antenna branches (102) for an iterative learning controlled ILC per branch MIMO DPD scheme with kernel regression; and

according to the ILC per branch MIMO DPD scheme with kernel regression, predistorting (202) a plurality of input signals for the plurality of antenna branches (102) respectively based on the determined one or more model parameters to provide a respective plurality of predistorted input signals for the plurality of antenna branches (102).

34. The method of claim 33, wherein determining the one or more model parameters for each antenna branch (102) of the plurality of antenna branches (102) comprises: for each of the antenna branches (102),

obtaining (300) N samples of a respective one of the plurality of input signals;

determining (302) an expected input signal U for the antenna branch (102)_l；

Generating (312) a kernel regression matrix θ for the antenna branch (102) based on the N samples of the respective one of the plurality of input signals_l(ii) a And

based on the kernel regression matrix theta for that antenna branch (102)_lAnd said desired input signal U_lCalculating (314) the one or more model parameters for the antenna branch (102).

35. The method as recited in claim 38, wherein generating (312) the kernel regression matrix Θ for the antenna branches (102)_lComprising generating (312) the kernel regression matrix θ for the antenna branches (102) based on_l: (a) the N samples of the respective one of the plurality of input signals and (b) N samples from one or more additional input signals of the plurality of input signals.

36. Method according to claim 34 or 35, wherein the desired input signal U for the antenna branch (102) is determined (302-310)_lThe method comprises the following steps:

initializing (302) the desired input signal U for the respective one of the plurality of antenna branches (102)_l；

Iteratively performing the following operations until at least one predefined criterion is met:

determining (306) an error E based on a difference between_l: (a) a desired output signal of said antenna branch (102) and (b) when said desired input signal U is present_lAn actual output signal of the antenna branch (102) when applied to the respective one of the plurality of antenna branches (102);

based on the error E_lUpdating (308) the desired input signal U for the antenna branch (102)_l。

37. The method of claim 34 or 36, wherein the one or more model parameters are weights applied to predistort the respective one of the plurality of input signals.

38. The method of any of claims 33 to 37, wherein the ILC per branch MIMO DPD scheme with kernel regression uses radial basis function RBF kernels.

39. The method of any of claims 33 to 38, wherein the ILC per branch MIMO DPD scheme with kernel regression takes into account memory effects on a predefined memory depth.

40. The method of any of claims 33 to 39, wherein the ILC per branch MIMO DPD scheme takes into account antenna coupling effects.

Technical Field

The present disclosure relates to Digital Predistortion (DPD) in a multiple-input multiple-output (MIMO) transmitter system.

Background

In Advanced Antenna Systems (AAS), also known as massive multiple-input multiple-output (MIMO) systems, the transmit signals are precoded to several (e.g., up to several hundred) transmit branches (also referred to herein as antenna branches) to enhance the achievable capacity through beamforming [1 ]. Each transmit branch is terminated by an antenna fed by a Power Amplifier (PA). As in conventional single-input single-output (SISO) systems, hardware imperfections such as the non-linear response of the PA cause distortion of the transmitted signal. Such distortion not only affects the transmitted signal itself, but also causes spectral broadening, which also damages adjacent channels. Furthermore, with AAS, having a large number of transmit branches means a large and complex hardware structure. Therefore, it is necessary to reduce hardware complexity. One way to reduce hardware complexity is to eliminate some Radio Frequency (RF) components and compensate for their absence using digital signal processing techniques. Such components include RF isolators, which are used in conventional systems to protect the PA from reflected back signals caused by antenna mismatch or antenna mutual coupling in multiple antenna systems [2 ]. Thus, unlike SISO systems or multi-antenna systems with isolators, in AAS, antenna mismatch and mutual coupling add more non-linear distortion, which also needs to be addressed when linearizing the PA by Digital Predistortion (DPD). The present disclosure relates to an isolator-free AAS or an AAS with relaxed isolation requirements. The AAS may be a Time Division Duplex (TDD) AAS or a Frequency Division Duplex (FDD) AAS.

There have been many studies on how to linearize the PA output in an isolator-less AAS by digitally pre-distorting the input signal. In [3-6], the output of a Dual Input (DI) PA (DI-PA) is modeled as a multivariate memory polynomial (M) that accounts for the front-to-back crosstalk of the PA. Motivated by the rapid complexity growth with increasing number of MIMO branches, in [7] a DI-PA model is proposed to model the PAs in an AAS with multiple branches, where for each DI-PA the first input is the corresponding PA direct input and the second input is the linear product between two vectors. The first vector is the output of the other branches and the second vector is the antenna array coupling coefficient between these branches and the respective branches. This DI-PA approach assumes that an antenna array coupling matrix is available, but it may not be available or at least needs to be estimated. U.S. patent application publication No. 2018/0167092 describes a MIMO transmitter that utilizes dual input digital predistortion (DI DPD) to compensate for a nonlinear PA. DI DPD may include the use of a transfer function based on radial basis functions.

One common point between the previously proposed linearization techniques for PAs in AAS is that they all use an Indirect Learning Architecture (ILA) to determine the parameters (e.g., weighting coefficients) of the predistorter [8 ]. However, in the case of highly correlated signals, ILA suffers from numerical problems due to the rank deficient system (rank specific system) that requires solving linear equations [9 ]. This high correlation between the different branch signals is exploited in [9], where the coupling between the highly correlated signals is described as simplifying the load variation of the DPD structure. However, the solution described in [9] does not apply when the signals are uncorrelated, so it is not a general solution. Therefore, for DPD in AAS, a low complexity, adaptive and general independent solution in terms of transmission rank (i.e. correlation) is needed.

Another common method in the previously proposed techniques for DPD in AAS is to have one DPD actuator per transmit branch, which means that both the extraction of DPD coefficients and DPD execution are performed as many times as the number of transmit branches in AAS. This means that the entire DPD system consumes a considerable amount of computational resources and hence power. Furthermore, in previously proposed techniques for DPD in AAS, the coupling between different signals before and after the PA is treated differently and knowledge about where the coupling occurs must be obtained. Therefore, there is also a need for a DPD solution for AAS that compromises performance, complexity and scalability. Furthermore, such a solution should also handle correlated and uncorrelated signals similarly. Furthermore, it is desirable that the solution handles coupling in the same way, regardless of whether the coupling occurs before or after the PA.

Some other notable DPD techniques are described in the following:

becerra et al, "effective linearization of RF transmitters by iterative Ridge Regression under a 5G waveform" (effective linearization of a RF Transmitter under 5G waveform) presents a method for digital predistortion of PAs using an algorithm, their so-called Iterative Ridge Regression (IRR), which operates on a Volterra kernel model,

U.S. patent application publication 2018/0167091 discloses a "massive" MIMO array that utilizes DPD to linearize PA circuits and discusses the advantages and disadvantages of using one shared DPD for all PAs or dedicated DPD for each PA, and

us patent application publication No. 2018/0316367 discloses a linearization method for antenna arrays (e.g. large arrays such as in AAS and "massive" MIMO). This reference teaches embodiments in which a single linearizer can be used to linearize all transceiver branches in an active antenna array, and embodiments in which the number of linearizers used to linearize a set of radio transceiver branches can be made significantly smaller than the number of radio transceiver branches in an AAS.

Disclosure of Invention

Systems and methods for Digital Predistortion (DPD) in a multiple-input multiple-output (MIMO) transmitter are disclosed herein. In some embodiments, a MIMO transmitter includes a plurality of antenna branches including a respective plurality of Power Amplifiers (PAs) coupled to a respective plurality of antenna elements. The MIMO transmitter also includes one or more DPD systems operable to determine, for each of one or more sets of antenna branches, one or more model parameters for a combined MIMO DPD scheme for the set of antenna branches, the combined MIMO DPD scheme being an Iterative Learning Controlled (ILC) combined MIMO DPD scheme with kernel regression. The one or more DPD systems are further operable to, for each of one or more sets of antenna branches, predistort a set of input signals for the set of antenna branches based on the determined one or more model parameters in accordance with a combined MIMO DPD scheme for the set of antenna branches to provide predistorted input signals for the respective set of the set of antenna branches. Each group of antenna branches includes at least two of the plurality of antenna branches. In some embodiments, the MIMO transmitter is a massive MIMO transmitter. By grouping the antenna branches, a combined or grouped combined MIMO DPD scheme is implemented, thereby achieving reduced computational complexity, reduced implementation complexity, and scalability.

In some embodiments, the one or more sets of antenna branches comprise two or more sets of antenna branches, the two or more sets of antenna branches being disjoint subsets of the plurality of antenna branches. Further, the one or more DPD systems include two or more DPD systems operable to predistort respective two or more sets of input signals to provide respective two or more sets of predistorted input signals for the two or more sets of antenna branches, respectively.

In some embodiments, each of the two or more DPD systems is further operable to acquire N samples of each of a plurality of input signals in a respective one of the two or more sets of input signals and determine a desired combined input signal U for the respective one of the two or more sets of antenna branches_s. The DPD system is further operable to generate a kernel regression matrix θ for a respective one of the two or more sets of antenna branches based on N samples in each of a plurality of input signals in the respective one of the two or more sets of input signals_sAnd based on a kernel regression matrix theta for a respective one of the two or more sets of antenna branches_sAnd the desired combined input signal U_sOne or more model parameters of a DPD model used by the DPD system are calculated to predistort a plurality of input signals in a corresponding one of the two or more sets of input signals. Further, in some embodiments, to determine a desired combined input signal U for a respective one of the two or more sets of antenna branches_sEach of the two or more DPD systems is further operable to initialize a desired combined input signal U for a respective one of the two or more sets of antenna branches_sAnd iteratively performing the following until at least one predefined criterion is met:

based on (a) the desired combined output signal of a respective one of the two or more sets of antenna branches and (b) the current desired combined input signalU_sDetermining an error E by a difference between actual combined output signals of a respective one of the two or more sets of antenna branches when applied to the respective one of the respective two or more sets of antenna branches_sAnd an

Based on the error E_sUpdating the desired combined input signal U for a respective one of the two or more sets of antenna branches_s。

In some embodiments, the one or more model parameters are weights applied by the DPD system to predistort a plurality of input signals in a respective one of the two or more sets of input signals.

In some embodiments, the one or more sets of antenna branches consists of a single set of antenna branches comprising multiple antenna branches, and the one or more DPD systems consists of a single DPD system operable to predistort the multiple input signals using an ILC combined MIMO DPD scheme for the multiple antenna branches with kernel regression to provide the multiple predistorted input signals for the multiple antenna branches. Further, in some embodiments, the single DPD system is further operable to obtain N samples of each of the plurality of input signals, determine a desired combined input signal U for the plurality of antenna branches, generate a kernel regression matrix θ for the plurality of antenna branches based on the N samples of each of the plurality of input signals, and calculate one or more model parameters of a DPD model used by the single DPD system based on the kernel regression matrix θ and the desired combined input signal U to predistort the plurality of input signals. Furthermore, in some embodiments, to determine the desired combined input signal U for the multiple antenna branches, the single DPD system is further operable to initialize the desired combined input signal U for the multiple antenna branches, and iteratively perform the following operations until at least one predefined criterion is met:

determining an error E based on a difference between (a) a desired combined output signal of the plurality of antenna branches and (b) an actual combined output signal of the plurality of antenna branches when applying a desired combined input signal U to the plurality of antenna branches, and

based on the error E, the desired combined input signal U for the multiple antenna branches is updated.

In some embodiments, the one or more model parameters are weights applied by a single DPD system to predistort multiple input signals.

In some embodiments, the ILC grouped MIMO DPD scheme with kernel regression or the ILC grouped MIMO DPD scheme with kernel regression for a respective one of two or more groups of antenna branches uses a Radial Basis Function (RBF) kernel.

In some embodiments, the ILC grouped MIMO DPD scheme with kernel regression or the ILC grouped MIMO DPD scheme with kernel regression for a respective one of two or more groups of antenna branches uses a Generalized Memory Polynomial (GMP) basis as the kernel.

In some embodiments, the ILC grouped MIMO DPD scheme with kernel regression or the ILC grouped MIMO DPD scheme with kernel regression for a respective one of two or more groups of antenna branches accounts for memory effects.

Embodiments of a method of performing DPD in a MIMO transmitter including a plurality of antenna branches including a respective plurality of power amplifiers coupled to a respective plurality of antenna elements are also disclosed. In some embodiments, a method of performing DPD in a MIMO transmitter comprises: for each of one or more groups of antenna branches, determining one or more model parameters for a combined MIMO DPD scheme for the group of antenna branches, wherein the combined MIMO DPD scheme is an ILC combined MIMO DPD scheme with kernel regression; and predistorting a set of input signals for the set of antenna branches based on the determined model parameters in accordance with a combined MIMO DPD scheme for the set of antenna branches to provide a corresponding set of predistorted input signals for the set of antenna branches. Each group of antenna branches includes at least two of the plurality of antenna branches.

In some embodiments, the one or more sets of antenna branches comprise two or more sets of antenna branches that are disjoint subsets of the plurality of antenna branches.

In some embodiments, determining the one or more model parameters of the combined MIMO DPD scheme for each of the two or more sets of antenna branches comprises: obtaining N samples of each of a plurality of input signals in the set of input signals; determining a desired combined input signal U for the set of antenna branches_s(ii) a Generating a kernel regression matrix θ for the set of antenna branches based on the N samples of each of the plurality of input signals in the set of input signals_s(ii) a And a kernel regression matrix theta based on the kernel regression matrix for the set of antenna branches_sAnd the desired combined input signal U_sOne or more model parameters for the combined MIMO DPD scheme for the set of antenna branches are calculated. In some embodiments, for each of two or more sets of antenna branches, a desired combined input signal U for the set of antenna branches is determined_sIncluding initializing a respective set of desired combined input signals U for a respective two or more sets of antenna branches_sAnd iteratively performing the following until at least one predefined criterion is met:

based on (a) the desired combined output signal of the set of antenna branches and (b) when applying the desired combined input signal U_sDetermining an error E by a difference between actual combined output signals of the group of antenna branches to a respective one of the respective two or more groups of antenna branches_sAnd an

Updating the desired combined input signal U for the set of antenna branches based on the error E_s。

Furthermore, in some embodiments, the one or more model parameters are weights applied to predistort a plurality of input signals in the set of input signals.

In some embodiments, the one or more sets of antenna branches consists of a single set of antenna branches including multiple antenna branches, and the combined MIMO DPD scheme for the set of antenna branches is an ILC combined MIMO DPD scheme for the multiple antenna branches with kernel regression. Further, in some embodiments, determining one or more model parameters of the combined MIMO DPD scheme for the single set of antenna branches includes: obtaining N samples for each of a plurality of input signals for the single set of antenna branches; determining a desired combined input signal U for a plurality of antenna branches in the single set of antenna branches; generating a kernel regression matrix θ for the plurality of antenna branches based on the N samples of each of the plurality of input signals; and calculating one or more model parameters for the ILC combined MIMO DPD scheme based on the kernel regression matrix θ and the desired combined input signal U. In some embodiments, for a single set of antenna branches, determining a desired combined input signal U for a plurality of antenna branches in the single set of antenna branches includes initializing the desired combined input signal for the plurality of antenna branches, and iteratively performing the following until at least one predefined criterion is met:

determining an error E based on a difference between (a) a desired combined output signal of the plurality of antenna branches and (b) an actual combined output signal of the plurality of antenna branches when applying the desired combined input signal U to the plurality of antenna branches, and

updating the desired combined input signal U for the plurality of antenna branches based on the error E when applying the desired combined input signal U to the plurality of antenna branches.

In some embodiments, the one or more model parameters for the ILC combined MIMO DPD scheme are weights applied to predistort the multiple input signals.

In some embodiments, the ILC grouped MIMO DPD scheme with kernel regression or the ILC grouped MIMO DPD scheme with kernel regression for a respective one of two or more groups of antenna branches uses RBF kernels.

In some embodiments, the ILC grouped MIMO DPD scheme with kernel regression or the ILC grouped MIMO DPD scheme with kernel regression for a respective one of two or more groups of antenna branches uses a GMP base as the kernel.

In some embodiments, the ILC grouped MIMO DPD scheme with kernel regression or the ILC grouped MIMO DPD scheme with kernel regression for a respective one of two or more groups of antenna branches takes into account memory effects.

In some other embodiments, a MIMO transmitter includes a plurality of antenna branches including a respective plurality of PAs coupled to a respective plurality of antenna elements, and a plurality of ILC DPD systems operable to individually predistort a plurality of input signals for the plurality of antenna branches using an ILC per branch MIMO DPD scheme with kernel regression.

In some embodiments, each of the plurality of ILC DPD systems is further operable to: obtaining N samples of a respective one of a plurality of input signals; determining a desired input signal U for a respective one of a plurality of antenna branches_l(ii) a Generating a kernel regression matrix θ for a respective one of the plurality of antenna branches based on the N samples of the respective one of the plurality of input signals_l(ii) a And kernel-based regression matrix theta_lAnd a desired input signal U for a respective one of the plurality of antenna branches_lOne or more model parameters for a DPD model utilized by the ILC DPD system are calculated to predistort a respective one of the plurality of input signals. In some embodiments, each of the plurality of ILC DPD systems is further operable to generate a kernel regression matrix θ for a respective one of the plurality of antenna branches based on (a) N samples of the respective one of the plurality of input signals and (b) N samples from one or more additional input signals of the plurality of input signals_l. In some embodiments, to determine the desired input signal U for a respective one of the plurality of antenna branches_lEach of the plurality of ILC DPD systems is further operable to initialize a desired input signal U for a respective one of the plurality of antenna branches_lAnd iteratively performing the following until at least one predefined criterion is met:

based on (a) the desired output signal of a respective one of the plurality of antenna branches and (b) when applying the desired input signal U_lDetermining an error E by a difference between actual output signals of a respective one of the plurality of antenna branches to the respective one of the plurality of antenna branches_lAnd an

Based on the error E_lUpdating the desired input signal U for a respective one of the plurality of antenna branches_l。

Each of the plurality of ILC DPD systems is further operable to: obtaining new N samples of a respective one of a plurality of input signals; generating a new kernel regression matrix θ for a respective one of the plurality of antenna branches based on the new N samples of the respective one of the plurality of input signals_l(ii) a And based on a new kernel regression matrix theta for a respective one of the plurality of antenna branches_lAnd a new desired input signal U_lOne or more new model parameter calculations are calculated for the DPD model used by the ILC DPD system to predistort a respective one of the plurality of input signals. In some embodiments, the one or more model parameters are weights applied by the ILC DPD system to predistort the respective one of the plurality of input signals.

In some embodiments, the ILC per branch MIMO DPD scheme with kernel regression uses RBF kernels.

In some embodiments, the ILC per branch MIMO DPD scheme with kernel regression takes into account memory effects of predefined memory depths.

In some embodiments, the ILC per branch MIMO DPD scheme takes into account antenna coupling effects.

In some other embodiments, a method of performing DPD in a MIMO transmitter including a plurality of antenna branches including a respective plurality of PAs coupled to a respective plurality of antenna elements includes: determining one or more model parameters for each antenna branch of a plurality of antenna branches for an ILC per branch MIMO DPD scheme with kernel regression; and predistorting the plurality of input signals for the plurality of antenna branches, respectively, based on the determined one or more model parameters in accordance with an ILC per branch MIMO DPD scheme with kernel regression to provide a corresponding plurality of predistorted input signals for the plurality of antenna branches.

In some embodiments, determining one or more model parameters for each of a plurality of antenna branches includes determining one or more model parameters for each of the plurality of antenna branches: for each antenna branch, taking N samples of a respective one of the plurality of input signals; determining a desired input signal U for the antenna branch_l(ii) a Generating a kernel regression matrix θ for the antenna branch based on the N samples of the respective one of the plurality of input signals_l(ii) a And kernel-based regression matrix theta_lAnd a desired input signal U for the antenna branch_lOne or more model parameters for the antenna branch are calculated. In some embodiments, a kernel regression matrix θ for the antenna branch is generated_lThe method comprises the following steps: based on: (a) n samples of a respective one of the plurality of input signals and (b) N samples from one or more additional input signals of the plurality of input signals, generating a kernel regression matrix θ for the antenna branch_l. In some embodiments, the desired input signal U for that antenna branch is determined_lThe method comprises the following steps: initializing a desired input signal U for a respective one of a plurality of antenna branches_lAnd iteratively performing the following operations until at least one predefined criterion is met:

based on (a) the desired output signal of the antenna branch and (b) the desired input signal U when applied_lDetermining an error E by the difference between the actual output signals of a plurality of antenna branches to a respective one of the antenna branches_lAnd an

Based on the error E_lUpdating the desired input signal U for the antenna branch_l。

Determining one or more model parameters for each of a plurality of antenna branches further comprises: for each antenna branch, taking new N samples of a respective one of a plurality of input signals; generating a new kernel regression matrix θ for the antenna branch based on the new N samples of the respective one of the plurality of input signals_l(ii) a And based on the new kernel regression matrix theta_lAnd a new desired input signal U for the antenna branch_lOne or more new model parameters for the antenna branch are calculated.

In some embodiments, the one or more model parameters are weights applied to predistort the respective one of the plurality of input signals.

In some embodiments, the ILC per branch MIMO DPD scheme with kernel regression uses RBF kernels.

In some embodiments, the ILC per branch MIMO DPD scheme with kernel regression takes into account memory effects of predefined memory depths.

In some embodiments, the ILC per branch MIMO DPD scheme takes into account antenna coupling effects.

Drawings

The accompanying drawings incorporated in and forming a part of the specification illustrate several aspects of the present disclosure, and together with the description serve to explain the principles of the disclosure.

Fig. 1 illustrates an example of a multiple-input multiple-output (MIMO) transmitter performing Digital Predistortion (DPD) according to an Iterative Learning Control (ILC) per-branch multiple-input single-output (MISO) DPD scheme, in accordance with some embodiments of the present disclosure;

fig. 2 is a flow chart illustrating operation of the MIMO transmitter of fig. 1 in accordance with some embodiments of the present disclosure;

fig. 3 is a flowchart illustrating operation of an ILC DPD system for the l-th antenna branch of the MIMO transmitter of fig. 1 according to some embodiments of the present disclosure;

fig. 4 shows one example of a set of antenna branches considered by the ILC DPD system for the l-th antenna branch according to one example embodiment of the present disclosure;

fig. 5 illustrates an example of a MIMO transmitter performing DPD according to an ILC combined MIMO DPD scheme, according to some embodiments of the present disclosure;

fig. 6 is a flow chart illustrating operation of the MIMO transmitter of fig. 5 in accordance with some embodiments of the present disclosure;

fig. 7 is a flowchart illustrating operation of the combined DPD system of the MIMO transmitter of fig. 5 according to some embodiments of the present disclosure;

figures 8 to 11 show simulation results for one example implementation of the combined MIMO DPD scheme used by the combined DPD systems described in figures 5 and 7;

fig. 12 illustrates an example of a MIMO transmitter performing DPD according to a packet-combined DPD scheme, according to some embodiments of the present disclosure;

fig. 13 is a flow chart illustrating operation of the MIMO transmitter of fig. 12 in accordance with some embodiments of the present disclosure;

fig. 14 is a flowchart illustrating operation of an s combination DPD system for an s set of antenna branches in the MIMO transmitter of fig. 12 according to some embodiments of the present disclosure;

figure 15 shows some example antenna branch groupings for a dual-polarized antenna array having 64 antenna elements per polarization;

figures 16 and 17 show simulation results for one particular implementation of packet-combining MIMO DPD; and

fig. 18 shows several MIMO DPD options.

Detailed Description

The embodiments set forth below represent information that enables those skilled in the art to practice the embodiments and illustrate the best mode of practicing the embodiments. Upon reading the following description in light of the accompanying drawing figures, those skilled in the art will understand the concepts of the disclosure and will recognize applications of these concepts not particularly addressed herein. It should be understood that these concepts and applications fall within the scope of the disclosure.

Systems and methods for Digital Predistortion (DPD) in a multiple-input multiple-output (MIMO) transmitter system are disclosed. Preferably, the MIMO transmitter is a massive MIMO transmitter system, also referred to herein as an Advanced Antenna System (AAS). As used herein, a massive MIMO transmitter system may have up to hundreds of transmit branches, or even more antenna branches. Furthermore, the MIMO transmitter system is preferably isolator-free (i.e., architecture with no Radio Frequency (RF) isolator). As described in detail below, in some embodiments, a MIMO transmitter system includes a DPD system that operates according to Iterative Learning Control (ILC) per-branch multiple-input-single-output (MISO) DPD with a kernel regression scheme. In some other embodiments, the MIMO transmitter system comprises a DPD system operating according to a combined or packet-combined MIMO DPD scheme. In some embodiments, the combined or packet-combined MIMO DPD scheme uses ILC. In some embodiments, the combined or grouped combined MIMO DPD scheme uses ILC with kernel regression.

In this regard, fig. 1 illustrates an example of a MIMO transmitter 100 that performs DPD according to the ILC per branch MISO DPD scheme, according to some embodiments of the present disclosure. The ILC per branch MISO DPD scheme uses kernel regression, as described below. As shown, MIMO transmitter 100 includes a plurality of antenna branches 102-1 through 102-L, where "L" is the number of antenna branches. The antenna branch 102 may alternatively be referred to herein as a "transmit branch" or a "radio branch". The antenna branches 102-1 to 102-L include respective digital-to-analog conversion (DAC) and up-conversion circuits 104-1 to 104-L and respective Power Amplifiers (PAs) 106-1 to 106-L. The outputs of the PAs 106-1 to 106-L are coupled to respective antenna elements 108-1 to 108-L.

MIMO transmitter 100 further comprises ILC DPD systems 110-1 to 110-L for respective antenna branches 102-1 to 102-L. The ILC DPD systems 110-1 to 110-L perform DPD in the MIMO transmitter 110 using kernel regression and ILC, as detailed below, rather than using multivariate memory polynomials as PA models in the MIMO transmitter 100 in the legacy mode and then finding the inverse model using the Indirect Learning Architecture (ILA). Although not shown, each ILC DPD system 110 includes a DPD actuator to predistort the respective input signals based on one or more DPD parameters (e.g., weights) and a DPD adapter to adaptively configure the DPD parameters based on the feedback signals from the output of the respective antenna branch 102 and optionally one or more input signals for one or more of the other antenna branches 102, as described in detail below. Here, the DPD adapter adaptively calculates DPD parameters using ILC with kernel regression, as will be described in detail below. Note that in the embodiments described herein, there are two DPD actuators, one for amplitude and one for phase. MIMO transmitter 100 further includes feedback branches 112-1 through 112-L having inputs coupled to outputs of respective antenna branches 102-1 through 102-L and outputs providing feedback signals to respective ILC DPD systems 110-1 through 110-L. Those skilled in the art will appreciate that feedback branches 112-1 through 112-L generally include a down-conversion circuit and an analog-to-digital conversion (ADC) circuit.

In operation, the respective antenna branches 102-1 to 102-input signal x of L₁To x_LIs pre-distorted by a respective ILC DPD system 110-1 to 110-L to provide a respective pre-distorted input signal u₁To u_L. Predistortion input signal u₁To u_LAnd then processed by the corresponding antenna branches 102-1 through 102-L to provide a transmit signal y₁To y_L。

Fig. 2 is a flow chart illustrating operation of MIMO transmitter 100 in accordance with some embodiments of the present disclosure. As shown, the ILC DPD systems 110-1 to 110-L use ILCs with kernel regression to determine one or more DPD model parameters for the respective antenna branches 102-1 to 102-L (step 200). Therefore, the MIMO transmitter 100 is said to perform MIMO DPD using the ILC per-branch DPD scheme with kernel regression. Furthermore, in some embodiments, each ILC DPD system 110 uses additional input signals for a subset of the other antenna branches 102 to account for coupling between antenna elements 108 when determining DPD model parameters. When doing so, the ILC per branch DPD scheme is referred to herein as the ILC per branch MISO DPD scheme. ILC DPD systems 110-1 through 110-L input signal x for respective antenna branches 102-1 through 102-L based on the determined DPD model parameters₁To x_LPredistortion is performed to provide respective predistorted input signals u for antenna branches 102-1 through 102-L₁To u_L(step 202).

Fig. 3 is a flowchart illustrating operation of the ILC DPD system 110-l for the l-th antenna branch 102-l according to some embodiments of the present disclosure. In particular, the process may be understood as a detail of one embodiment of step 200 of FIG. 2. Note that this process is performed separately by each of the L ILC DPD systems 110-1 through 110-L. The description herein is for one of those ILC DPD systems, which is labeled as the ith ILC DPD system 110-1. It is further noted that where the term "step" is used, unless otherwise required, the "steps" (or actions) may be performed in any suitable order, and in some implementations, some steps may be performed concurrently.

As shown, the ILC DPD system 110-l for the l-th antenna branch 102-l acquires an input signal x for the l-th antenna branch 102-l_lN input samples (step 300). The set of input samples may be expressed in matrix form as:

as shown, the ILC DPD system 110-l determines the desired PA input signal (U) for the l-th antenna branch 102-l_l) (step 302). Here, in this context, this may also be said to initialize the desired PA input signal (U) for the l-th antenna branch 102-l_l). In this example, a PA input signal (U) is desired_l) Is initialized to:

wherein, U_lIs defined as:

note that u_l(n) denotes the desired input signal u for the l-th antenna branch 102-l_lThe nth sample of the (i.e., the predistorted input signal), and N is the total number of samples. Y is_l,dIs the desired combined output signal of the l-th antenna branch 102-l. In other words, Y_l,dIs a linearly amplified output signal, which can be expressed as:

Y_l，d＝gX_l。

ILC DPD system 110-1 calculates the desired PA input signal (U)_l) Amplitude component (U) of_l,amp) And a phase component (U)_l,phase) (step 304). Desired PA input signal (U)_l) Amplitude component (U) of_l,amp) And a phase component (U)_l,phase) Can be expressed as:

and

note that, here, | u_l(n) | denotes the input signal u_lAnd phi (u) is the amplitude component of the nth sample of (1)_l(n)) denotes the input signal u_lThe phase component of the nth sample.

Although not shown in the flowchart of FIG. 3, the ILC DPD system 110-l would then expect a PA input signal (U)_l) Applied to antenna branch 102-1. Using the feedback signal from feedback branch 112-l, ILC DPD system 110-l measures the error (E) between_l): (a) desired PA input signal (U) generated from the l-th antenna branch 102-l_l) I.e. the output of the l-th antenna branch 102-l, is the actual output y of the PA 106-l (i.e. the output of the l-th antenna branch 102-l)_lAnd (b) the desired output signal y of the PA 106-l (i.e., the output of the 1 st antenna branch 102-1)_l,dN samples (step 306). In matrix form, the actual output signal y_lIs expressed as Y_lAnd the output signal y is expected_l,dIs expressed as Y_l,d。

ILC DPD System 110-l is based on measured error (E)_l) Calculating a new desired input signal U for the l-th antenna branch 102-l_l,new(step 308). In other words, ILD DPD system 110-l is based on the error (E) of the measurement_l) Updating the desired input signal U of the l-th antenna branch 102-l_l. In this example, a new (i.e., updated) desired input signal U_l,newIs calculated as follows:

U_l，new＝U_l+εE_l。

note that ε is a constant, called the learning gain, which is equivalent to 1/g, where "g" is the linear gain. Although not required, further explanation of ε can be found in [10 ].

The process then returns to step 304 and repeats, for example, until a desired convergence criterion is met (e.g., repeats until the error E is met)_lWithin a predefined acceptable rangeUntil a defined maximum number of iterations has been performed, etc.) (step 310).

Once convergence is reached, at this point a new (i.e., updated) desired input signal U for the l-th antenna branch 102-l_l,newIs determined as the desired input signal for the l-th antenna branch 102-l. This new desired input signal U of the l-th antenna branch 102-l at the convergence point is thus obtained_l,newDesired input signal U referred to as the l-th antenna branch 102-l_lA prompter for use in the process of fig. 3.

The ILC DPD system 110-l generates a kernel regression matrix θ for the l-th antenna branch 102-l_l(step 312). In example embodiments herein, the kernel regression matrix θ is generated in a manner that compensates for memory effects and coupling between the antenna elements 108-l and subsets of other antenna elements 108_l. More specifically, ILC DPD system 110-l generates kernel regression matrix θ as follows_l：

Will vector z_l(n) is defined as:

where J is the set of antenna branches for which mutual coupling is considered for the l-th antenna branch, where J is defined as:

J＝{j1，…，jC}

where C is the number of antenna branches for which mutual coupling is considered for the l-th antenna branch, j1 is the index of the first such antenna branch, j2 is the index of the second such antenna branch, jC is the index of the C-th such antenna branch. As an example, antenna branch set J includes adjacent antenna elements as shown in fig. 4.

Regression quantityIs defined as:

wherein

Note that kernel regression is based on the following assumptions: each input-output pair in the data set (or a representative centroid of the input space) affects an estimate of the nonlinear output corresponding to a particular input based on the euclidean distance of the two inputs in their vector space. The representative centroid of the input space is denoted as μ_kK:1 → K, where K is the number of representative centroids of the input space. In example embodiments described herein, the Loaded (Lloyd) algorithm may be used to identify the centroid μ_kK:1 → K. Since PA is an amplitude driving device [13]]The signal amplitude support is treated as input space and then the amplitude and phase responses are estimated separately. Note also that z_l,i(n) the index vector z_lThe ith element of (n).

In this example, a Radial Basis Function (RBF) kernel is used. Note, however, that any other type of core may be used (e.g., based on a Generalized Memory Polynomial (GMP) -based core). Using an RBF kernel, for each value of i 1,2, … L (M +1), where M is the memory depth considered for the memory effect of the L-th antenna branch 102-L, for each value of N (where N1,...., N) and K (where K1,...., K), the i-th regression for the L-th antenna branch 102-L is usedOf (2) element(s)Is calculated as follows:

importantly, i is the vector z_l(n) such that:

z_l，1(n)＝|x_l(n)|

z₂(n)＝|x_l(n-1)|

...

...

regression matrix θ of RBF kernel of the l-th antenna branch 102-l_lThe structure is as follows:

ILC DPD system 110-1 is based on kernel regression matrix theta_lAnd a desired input signal (U)_l) Amplitude component (U) of_l,amp) And a phase component (U)_l,phase) The DPD parameters for the ILC DPD system 110-1 are calculated (step 314). In this example, the DPD parameter is the amplitude weight (W) of the l-th antenna branch 102-l_l,amp) And phase weight (W)_l,phase) It is calculated as follows:

and

although not shown in the flow chart of FIG. 3, the ILC DPD system 110-l then applies the calculated weight W_l,ampAnd W_l,phaseTo input signal x_l(future samples of) predistorting to provide a predistorted input signal u_lWhich is then processed by the l-th antenna branch 102-l to output a transmit signal for the l-th antenna branch 102-ly_l。

Note that the process of FIG. 3 finds U_lAnd thus gives a corresponding desired pre-distorted input signal for respective samples of the input signal used to linearize the PA 106-l. The computational cost of performing linearization at a per sample level is too high; instead, U is modeled as a function of the input samples, which is referred to herein as "ILC DPD". It should also be noted that while the process of fig. 3 considers coupling between the antenna elements 108, the disclosure is not so limited. For example, the process may not consider coupling between the antenna branches 102, which is equivalent to J being an empty set.

Using the ILC per-branch DPD scheme described above, performance similar to the Dual Input (DI) PA (DI-PA) approach proposed in [7] can be achieved, however a five-fold reduction in floating point operations (FLOPS) is required. Technique [7] was chosen for this comparison, since it is, to the best knowledge of the inventors, the least complex MIMO DPD found in the literature. Complexity analysis was performed considering both DPD adaptation and DPD execution in a closed-loop setup.

Furthermore, the ILC per branch DPD scheme described above works in the same way for both correlated and uncorrelated signals. This provides an advantage over conventional DPD schemes which are valid for one but fail for the other (see, e.g., [9] DPD scheme applicable in the case of correlated signals, which would otherwise fail, and ILA based techniques fail when the signals are correlated). The correlation here is captured by the number of MIMO layers.

In the discussion above, MIMO DPD is performed by using a per-branch ILC DPD system. Now, the discussion will turn to some other embodiments of the present disclosure, wherein a combined or group-combined MIMO DPD scheme is provided.

In some embodiments of combined MIMO DPD, multiple output kernel regression and ILC are used to perform DPD in a combined or group-wise manner in a MIMO transmitter system (e.g., AAS) as an alternative to using multivariate memory polynomials as PA models in AAS and then using ILA to find the inverse model. In a per-branch DPD actuator setup, the ILC is used to identify the best predistorted input signal for each individual PA, which drives the PA output as a linear (or near linear) function of the input signal. Using combined or grouped combined MIMO DPD according to some embodiments of the present disclosure, input signals of all antenna branches are predistorted in a combined manner and the resulting predistorted input signals are transmitted to the respective antenna branches. The input signals of all antenna branches are collected into a combined input signal matrix X. Using ILC and kernel regression, the combined DPD model is configured to produce a desired predistorted input signal matrix U that drives all antenna branches to produce an output signal matrix Y that is a linear (or nearly linear, e.g., as linear as possible) amplified version of X.

Using the combined MIMO DPD scheme described herein provides many other advantages over other approaches. Compared to other solutions, the combined MIMO DPD scheme described herein has the following advantages:

the combined MIMO DPD scheme described herein provides reduced computational and implementation complexity compared to per-branch DPD actuator solutions, and

the combined MIMO DPD scheme described herein is independent of aspects such as correlation between signals, and where radio chain signals are coupled to each other. Therefore, less information is required to implement the combined MIMO DPD scheme. In fact, it is not even necessary to identify the non-linear order in advance. Only the required memory depth is required.

In this regard, fig. 5 illustrates an example of a MIMO transmitter 500 that performs DPD according to an ILC combined MIMO DPD scheme according to some embodiments of the present disclosure. As described below, in some embodiments, the ILC combined MIMO DPD scheme uses kernel regression. As shown, MIMO transmitter 500 includes a plurality of antenna branches 502-1 through 502-L, where "L" is the number of antenna branches. The antenna branch 502 may alternatively be referred to herein as a "transmit branch" or "radio branch". The antenna branches 502-1 to 502-L include respective DAC and upconversion circuits 504-1 to 504-L and respective PAs 506-1 to 506-L. The outputs of the PAs 506-1 through 506-L are coupled to respective antenna elements 508-1 through 508-L.

MIMO transmitter 500 also includes an ILC combined MIMO DPD system 510 (also referred to herein as "combined DPD system 510"). Combined DPD System 510 uses ILC and in some instancesKernel regression is used in embodiments to perform DPD in MIMO transmitter 510, as described in detail below. Although not shown, the combined DPD system 510 includes a combined DPD actuator and DPD adapter, the combined DPD actuator combining the input signals x for the respective antenna branches 502-1 to 502-L based on one or more DPD parameters (e.g., weights) of the combined MIMO DPD model for the MIMO transmitter 500₁To x_LWith predistortion, the DPD adapter adaptively configures the DPD parameters based on the feedback signals from the outputs of the antenna branches 502-1 to 502-L, as described in detail below. Herein, the DPD adapter adaptively calculates DPD parameters using ILC, preferably with kernel regression, as will be described in detail below. Note that in the embodiments described herein, there are two DPD actuators, one for amplitude and one for phase. MIMO transmitter 500 further includes feedback branches 512-1 through 512-L having inputs coupled to the outputs of respective antenna branches 502-1 through 502-L and an output providing a feedback signal to combined DPD system 110. As will be appreciated by those skilled in the art, the feedback branches 512-1 to 512-L generally include down-conversion circuitry and ADC circuitry.

In operation, input signal x for respective antenna branches 502-1 to 502-L₁To x_LPre-distortion by combined DPD system 510 based on a combined MIMO DPD model to provide a corresponding pre-distorted input signal u₁To u_L. Predistortion input signal u₁To u_LAnd then processed by the corresponding antenna branches 502-1 through 502-L to provide a transmit signal y₁To y_L。

Fig. 6 is a flow chart illustrating operation of MIMO transmitter 500 in accordance with some embodiments of the present disclosure. As shown, the combined DPD system 510 determines one or more DPD model parameters (e.g., using ILC with kernel regression) for the combined MIMO DPD model for the L antenna branches 502-1 to 502-L (step 600). Thus, MIMO transmitter 500 is said to perform combined MIMO DPD, e.g., using ILC with kernel regression. Combining the input signals x of the DPD system 510 to the respective antenna branches 502-1 to 502-L₁To x_LPerforming pre-distortion to antenna diversity based on the determined DPD model parameters for combining MIMO DPD modelsThe branches 502-1 to 502-L provide respective predistorted input signals u₁To u_L(step 602).

Fig. 7 is a flowchart illustrating the operation of the combined DPD system 510 according to some embodiments of the present disclosure. In particular, the process may be understood as a detail of one embodiment of step 600 of FIG. 6. Note that where the term "step" is used, these "steps" (or actions) may be performed in any suitable order, unless otherwise required, and in some embodiments some steps may be performed simultaneously.

As shown, combined DPD system 510 takes input signal x for L antenna branches 502-1 through 502-L₁To x_LN input samples of each (step 700). This set of input samples can be expressed as an input signal matrix X, as follows:

wherein x is_l(n) denotes the input signal x for the l-th antenna branch 502-l_lThe nth sample of (1).

As shown, combined DPD system 510 determines the desired combined input signal (U) for L antenna branches 502-1 through 502-L (step 702). In other words, combined DPD system 510 initializes the desired combined input signal (U) for the L antenna branches 502-1 through 502-L. In this example, the desired combined input signal (U) is calculated as follows:

wherein U is defined as:

note that u_l(n) indicates the desired input signal u for the l-th antenna branch 502-l_l(i.e., predistorting the input signal) and N is for each antenna branch 502Total number of samples. Y is_dIs the desired combined output signal for antenna branch L antenna branches 502-1 through 502-L. In other words, Y_dIs a linearly amplified output signal, which can be expressed as:

Y_d＝gX。

the combined DPD system 510 calculates the amplitude component (U) of the desired combined input signal (U)_amp) And a phase component (U)_phase) (step 704). Desirably combining the amplitude components (U) of the input signal (U)_amp) And a phase component (U)_phase) Can be expressed as:

and

note that, here, | u_l(n) | denotes the input signal u of the l-th antenna branch 502-l_lAnd phi (u) is the amplitude component of the nth sample of (1)_l(n)) denotes the input signal u of the l-th antenna branch 502-l_lThe phase component of the nth sample.

Although not shown in the flow chart of fig. 7, combined DPD system 510 then applies the desired combined input signal (U) to antenna branches 502-1 to 502-L. Using the feedback signals from the feedback branches 512-1 to 512-L, the combined DPD system 510 measures the error (E) between: (a) actual output signals y of antenna branches 502-1 to 502-L resulting from desired combination of N samples of input signal (U)₁To y_L(represented as the actual output signal matrix Y) and (b) the desired output signals Y of the antenna branches 502-1 to 502-L_1,dTo y_L,d(expressed as the actual output signal matrix Y_d) (step 706).

Combined DPD system 510 calculates a new desired combined input signal U based on the measured error (E)_new(step 708). In other words, combined DPD system 510 updates the desired combined input signal U based on the measured error (E). Therein, theIn an example, a new (i.e., updated) desired input signal U_newIs calculated as follows:

U_new＝U+E。

note that ε is a constant, called the learning gain, which is equivalent to 1/g, where "g" is the linear gain. Although not required, further explanation of ε can be found in [10 ].

The process then returns to step 704 and repeats, for example, until a desired convergence criterion is met (e.g., until the error E is within a predefined acceptable range, until a defined maximum number of iterations has been performed, etc.) (step 710).

Once convergence is reached, a new desired combined input signal U at that point_newIs determined as the desired combined input signal. Thus, this new desired input signal U at the convergence point_newReferred to as the desired combined input signal U, for the reminder of the process of fig. 7.

The combined DPD system 510 generates a kernel regression matrix θ for the L antenna branches 502-1 through 502-L (step 712). In the example embodiments herein, the kernel regression matrix θ is generated in a manner that compensates for memory effects. To introduce memory effects, the tapped delay inputs are merged into the regression matrix θ by concatenating the regressors for the current and past inputs up to a predefined memory depth M, as described below. More specifically, combined DPD system 510 generates kernel regression matrix θ as follows:

define the vector z (n) as:

where M is the memory depth used to generate the memory effect considered by the kernel regression matrix θ.

Regression quantityIs defined as:

wherein

Note that kernel regression is based on the following assumptions: each input-output pair in the data set (or a representative centroid of the input space) affects an estimate of the nonlinear output corresponding to a particular input based on the euclidean distance of the two inputs in their vector space. The representative centroid of the input space is denoted as μ_kK:1 → K, where K is the number of representative centroids of the input space. In example embodiments described herein, the Loadeld algorithm may be used to identify the centroid μ_kK:1 → K. Since PA is an amplitude driving device [13]]The signal amplitude support is treated as input space and then the amplitude and phase responses are estimated separately. Note also that z_i(n) denotes the i-th element of the vector z (n).

In this example, an RBF kernel is used. Note, however, that any other type of core (e.g., GMP-based) may be used. Using an RBF kernel, for each value of i 1,2, … L (M +1), for each value of N (where N1.., N) and K (where K1.., K) for the ith regression θⁱOf (2) element(s)Is calculated as follows:

importantly, i is the index of the vector z (n), such that:

z₁(n)＝|x₁(n)|

z₂(n)＝|x₁(n-1)|

...

z_M+1(n)＝|x₁(n-M)|

...

z_L(M+1)(n)＝|x_L(n-M)|。

constructing the RBF kernel regression matrix θ as:

the combined DPD system 510 combines the amplitude components (U) of the input signal (U) based on the kernel regression matrix θ and the desired combined input signal (U)_amp) And a phase component (U)_phase) DPD parameters of the combined MIMO DPD model are calculated (step 714). In this example, the DPD parameter is an amplitude weight (W)_amp) And phase weight (W)_phase) They are calculated using a multiple output regression as:

W_amp＝(θ^Tθ)^-1θ^TU_amp

and

W_phase＝(θ^Tθ)^-1θ^TU_phase。

wherein, the total number of parameters (amplitude and phase) used for the whole combined MIMO DPD model is:

the # parameter is 2((1+ LK (M +1)) × L).

Note that W_ampAre labeled as:

·w_1,lwhich is the amplitude offset term for the l-th antenna branch 502-l,

·w_1+l(M+1),lwhich is the amplitude response directly input to the antenna branch i, and

·W_ampis the magnitude response to the current and past values of the different antenna branch inputs.

Likewise, W_phaseAre labeled as:

·w_1,lwhich is the phase offset term for the l-th antenna branch 502-l,

·w_1+l（M+1),lwhich is the phase response directly input to the antenna branch i, and

·W_phaseis the phase response to the current and past values of the different antenna branch inputs.

Although not shown in the flow chart of fig. 7, combined DPD system 510 is then applied to input signal x₁To x_LApplying the calculated weight W to the samples of each of (1) when pre-distorting_ampAnd W_phaseTo provide a pre-distorted input signal u₁To u_LWhich is then processed by the respective antenna branches 502-1 to 502-L to output a transmit signal y₁To y_L。

Note that the process of fig. 7 finds U and thus gives the corresponding desired pre-distorted input signal for the samples of the input signal that linearize the PAs 506-1 through 506-L. The computational cost of performing linearization at a per sample level is too high; instead, U is modeled as a function of the input samples, which is referred to herein as "ILC combined DPD".

Note that for the kernel decay parameter γ_kK:1 → K, optimization, Expectation Maximization (EM) may be used to jointly optimize these attenuation parameters as well as the amplitude and phase weights. Gradient descent for updating gamma_kK:1 → K. Further details and further explanation regarding this problem may be found in [14]]Is found in (1).

Fig. 8-11 show simulation results for one example implementation of the combined MIMO DPD scheme described above with respect to fig. 5 and 7. These simulation results show DPD performance in terms of Adjacent Channel Leakage Ratio (ACLR), Error Vector Magnitude (EVM) and number of FLOPS.

Simulation results show that:

performance: the smaller the antenna array size L, the better the combined MIMO DPD performance measured in ACLR and EVM. The reason for this is that less coupling is introduced in smaller arrays and regression matrices with smaller dimensions are used for kernel regression.

Complexity: the complexity of combined MIMO DPD measured in FLOPS increases dramatically as the antenna array size L increases.

Furthermore, compared to the DI-PA GMP based DPD introduced in [7], the following comments can be made regarding implementation and scalability:

embodiment (c): for combined MIMO DPD using ILC and kernel regression, there are always two DPD actuators (one for amplitude and one for phase). For DI-PA GMP, on the other hand, the number of DPD actuators is equal to the number of branches L.

Extensibility: DI-PA GMP is more scalable than combined MIMO-DPD using ILC and kernel regression.

Next, an "in-between solution" is proposed herein, in which not all antenna branches are combined, but also each branch DPD is not used. This "intermediate solution" is referred to herein as grouped combined MIMO DPD, wherein the antenna branches are grouped into groups (preferably, the number of groups is greater than 1 and less than L), wherein each group of antenna branches has its own DPD system. In some embodiments, for each group of antenna branches, the respective DPD system uses ILC and kernel regression, as described in detail below.

In this regard, fig. 12 illustrates an example of a MIMO transmitter 1200 that performs DPD according to a packet-combined DPD scheme according to some embodiments of the present disclosure. As described below, in some embodiments, the packet-combined MIMO DPD scheme uses ILC and possibly kernel regression to perform combined MIMO DPD for each of multiple sets of antenna branches. As shown, MIMO transmitter 1200 includes multiple (L) antenna branches. The antenna branches may alternatively be referred to herein as "transmit branches" or "radio branches". The antenna branches are grouped into a plurality of groups or sets. In this example, each group has the same number of antenna branches, denoted by S. Note, however, that this is only one example. Different groups may include different numbers of antenna branches. When the number of groups of antenna branches is greater than 1, then the groups are disjoint subsets of antenna branches. Preferably, the number of groups (L/S) is greater than 1 and less than L.

For clarity, the antenna branches are represented by respective PAs 1202- (1, 1) to 1202- (L/S, S). Note, however, that the antenna branches comprise additional circuitry, such as, for example, respective DACs and up-conversion circuitry. The outputs of the PAs 1202- (1, 1) through 1202- (L/S, S) are coupled to respective antenna elements 1204- (1, 1) through 1204- (L/S, S). Note that the reference notation of a PA (and corresponding input signal, output signal, and antenna) uses the notation (a, B), where "a" is the index of the group and "B" is the index of the PA/input signal/output signal within the group.

MIMO transmitter 1200 also includes combined MIMO DPD systems 1206-1 to 1206- (L/S) for respective antenna branch groups. In other words, MIMO transmitter 1200 includes a combined MIMO DPD system 1206 for each group of antenna branches that performs combined MIMO DPD for that group of antenna branches. And each combined MIMO DPD system 1206 may utilize any desired combined MIMO DPD scheme. In some embodiments, each combined MIMO DPD system 1206 uses ILC, and in some embodiments uses kernel regression to perform combined DPD for the respective set of antenna branches, as described in detail below. Although not shown, for each S-th group of antenna branches (where S:1 → L/S), the S-th combined MIMO DPD system 1206-S includes a combined DPD actuator and a DPD adaptor, the combined DPD actuator combinatorially applying the input signals x for the respective groups of antenna branches to the input signals x for the respective groups of antenna branches based on one or more DPD parameters (e.g., weights) of a combined MIMO DPD model for the respective groups of antenna branches_s,1To x_s,SWith predistortion, the DPD adapter adaptively configures the DPD parameters based on the output feedback signals from the respective groups of antenna branches, as described in detail below. In some embodiments, the DPD adapter adaptively calculates DPD parameters using ILC, preferably with kernel regression, as will be described in detail below. Note that in the embodiments described herein, there are two DPD actuators, one for amplitude and one for phase, in the s-th DPD combined MIMO DPD system 1206-s. MIMO transmitter 1200 also includes feedback branches (not shown for clarity) having inputs coupled to the outputs of the respective antenna branches and outputs providing feedback signals to respective combined MIMO DPD systems 1206. As will be understood by those skilled in the art, the feedback branch generally includes a down-conversion circuit and an ADC circuit.

In operation, for each S-th group antenna branch (where S:1 → L/S), the input signal x for the S-th group antenna branch_s,1To x_s,SCombining MIMO DPD systems 1206-s by sPredistortion in a combined MIMO DPD model for the s-th group of antenna branches to provide a corresponding predistorted input signal u_s,1To u_s,S. Predistortion input signal u_s,1To u_s,SAnd then processed by the corresponding antenna branches (as represented in fig. 12 by the corresponding PAs 1202- (S, 1) to 1202- (S, S)) to provide the transmitted signal y to the S-th group of antenna branches_s,1To y_s,S。

Fig. 13 is a flow chart illustrating operation of MIMO transmitter 1200 in accordance with some embodiments of the present disclosure. As shown, for each s-th group of antenna branches, the combined MIMO DPD system 1206-s for the s-th group of antenna branches (e.g., using ILC with kernel regression) determines one or more DPD model parameters for the combined MIMO DPD model for the s-th group of antenna branches (step 1300). Thus, MIMO transmitter 1200 is referred to as performing packet-combining MIMO DPD using ILC with kernel regression, for example. Combined MIMO DPD system 1206-s for group s antenna branches pairs input signals x for group s antenna branches using a combined DPD model for group s antenna branches and determined DPD parameters_s,1To x_s,SPredistortion to provide a predistorted input signal u for the s-th group of antenna branches_s,1To u_s,S(step 1302).

Fig. 14 is a flow chart illustrating operation of an s-th combined MIMO DPD system 1206-s for an s-th group of antenna branches in MIMO transmitter 1200 of fig. 12 according to some embodiments of the present disclosure. In particular, the process may be understood as a detail of one embodiment of step 1300 of FIG. 13. Note that where the term "step" is used, these "steps" (or actions) may be performed in any suitable order, unless otherwise required, and in some embodiments some steps may be performed simultaneously.

As shown, a combined MIMO DPD system 1206-S for an S-th group of antenna branches obtains input signals x for S antenna branches of the S-th group of antenna branches_s,1To x_s,SN input samples of each (step 1400). The set of input samples may be expressed as an input signal matrix X_sAs follows:

wherein x is_s,1(n) denotes the input signal x of the first antenna branch in the s-th group of antenna branches_s.1N-th sample of (1), x_s,2(n) denotes the input signal x of the second antenna branch of the s-th group of antenna branches_s.2The nth sample of (a), and so on.

Combined MIMO DPD system 1206-s for group s antenna branches determines a desired combined input signal (U) for group s antenna branches_s) (step 1402). In other words, the combined MIMO DPD system 1206-s for the s-th group of antenna branches initializes the desired combined input signals (U) for the s-th group of antenna branches_s). In this example, the desired combined input signal (U) for the s-th group of antenna branches_s) Is calculated as follows:

wherein, U_sIs defined as:

note that u_s,1(n) identifying the desired input signal u for the first antenna branch in the s-th group of antenna branches_s,1(i.e. the predistorted input signal) of the nth sample, u_s,2(n) denotes a desired input signal u for a second antenna branch of the s-th group of antenna branches_s,2The nth sample of the (i.e., the predistorted input signal), and so on. Further, N is the total number of samples for each antenna branch in the s-th group of antenna branches. Y is_s,dIs the desired combined output signal for the s-th group of antenna branches. In other words, Y_s,dIs a linearly amplified output signal that can be expressed as:

Y_s，d＝gX_s。

combined MIMO DPD system 1206-s for group s antenna branchesDesired combined input signal (U) at the s-th antenna branch_s) Amplitude component (U) of_s,amp) And a phase component (U)_s,phase) (step 1404). Desired combined input signal (U) for the s-th group of antenna branches_s) Amplitude component (U) of_s,amp) And a phase component (U)_s,phase) Can be expressed as:

and

note that, here, | u_s,1(n) | denotes the input signal u for the first antenna branch of the s-th group of antenna branches_s,1The amplitude component, | u, of the nth sample of (1)_s,2(n) | denotes the input signal u for the second antenna branch of the s-th group of antenna branches_s,2The amplitude component of the nth sample, and so on. Likewise, phi (u)_s,1(n)) denotes the input signal u of the first antenna branch of the s-th group of antenna branches_s,1The phase component of the nth sample, phi (u)_s,2(n)) denotes the input signal u of the second antenna branch of the s-th group of antenna branches_s,2The phase component of the nth sample, and so on.

Although not shown in the flow chart of fig. 14, the combined MIMO DPD system 1206-s for the s-th set of antenna branches would then expect to combine the input signals (U)_s) To the s-th group of antenna branches. Combining MIMO DPD systems 1206-s using feedback signals from feedback paths of the s-group antenna branches measures the error (E) between_s): (a) combining input signals (U) by expectation_s) Resulting in an actual output signal y for an antenna branch of the s-th group of antenna branches_s,1To y_s,S(expressed as the actual output signal matrix Y_s) And (b) desired output signal y for the s-th group of antenna branches_s,1,dTo y_s,S,d(expressed as the actual output signal matrix Y_s,d) (step 1406).

Combined MIMO DPD system 1206-s for group s antenna branches based on measured error (E) of group s antenna branches_s) Calculating a new expected combined input signal U for the s-th group of antenna branches_s,new(step 1408). In other words, the combined MIMO DPD system 1206-s for the s-th group of antenna branches is based on the measured error (E) of the s-th group of antenna branches_s) The desired combined input signal Us for the s-th group of antenna branches is updated. In this example, a new (i.e., updated) desired input signal U for the s-th group of antenna branches_newThe calculation is as follows:

U_s，new＝U_s+εE_s。

note that ε is a constant, called the learning gain, which is equivalent to 1/g, where "g" is the linear gain. Although not required, further explanation of ε can be found in [10 ].

The process then returns to step 1404 and repeats, for example, until a desired convergence criterion is met (e.g., until error E is reached_sRepeat until a defined maximum number of iterations has been performed, within a predefined acceptable range, and so on) (step 1410).

Once convergence is reached, a new desired combined input signal U for the s-th group of antenna branches at that point_s,newAnd then determined as the desired combined input signal for the s-th group of antenna branches. This new desired combined input signal U for the s-th group of antenna branches at the convergence point is thus obtained_s,newDesired combined input signal U, referred to as s-th group of antenna branches_sA reminder for the process of fig. 14.

Combined MIMO DPD system 1206-S for the S-th group of antenna branches generates a kernel regression matrix θ for the S antenna branches in the S-th group of antenna branches_s(step 1412). In example embodiments herein, the kernel regression matrix θ is generated in a manner that compensates for memory effects_s. To introduce memory effects, the tapped delay inputs are merged into the regression matrix θ by concatenating the regressors for the current and past inputs up to a predefined memory depth M, as described below. More specifically, the present invention is to provide a novel,combined MIMO DPD system 1206-s for the s-th group of antenna branches generates a kernel regression matrix θ as follows_s：

Will vector z_s(n) is defined as:

where M is used to generate the kernel regression matrix θ_sMemory depth for memory effects to be considered.

Regression quantityIs defined as:

wherein

Note that kernel regression is based on the following assumptions: each input-output pair in the data set (or a representative centroid of the input space) affects an estimate of the nonlinear output corresponding to a particular input based on the euclidean distance of the two inputs in their vector space. The representative centroid of the input space is denoted as μ_kK:1 → K, where K is the number of representative centroids of the input space. In example embodiments described herein, the Loadeld algorithm may be used to identify the centroid μ_kK:1 → K. Since PA is an amplitude driving device [14]]The signal amplitude support is treated as input space and then the amplitude and phase responses are estimated separately. Note also that z_s,i(n) the index vector z_sThe ith element of (n).

In this example, an RBF kernel is used. Note, however, that any other type of core (e.g., GMP-based) may be used. Using RBF kernel, for each value of i 1,2, … S (M +1), for each value of nEach of the values (where N is 1,.. cndot., N) and K (where K is 1.. cndot., K) is used for the ith regression quantityOf (2) element(s)Is calculated as follows:

importantly, i is the index of the vector z (n), such that:

z_s，1(n)＝|x_s，1(n)|

z_s，2(n)＝|x_s，1(n-1)|

...

z_s，M+1(n)＝|x_s，1(n-M)|

...

z_S(M+1)(n)＝|x_s，S(n-M)|。

regression matrix θ with RBF kernel_sThe structure is as follows:

combined MIMO DPD system 1206-s for group s antenna branches based on kernel regression matrix theta_sAnd a desired combined input signal (U) for the s-th group of antenna branches_s) Amplitude component (U) of_s,amp) And a phase component (U)_s,phase) DPD parameters for the combined MIMO DPD model for the s-th set of antenna branches are calculated (step 1414). In this example, the DPD parameter is a magnitude weight (W) calculated using multi-output regression_s,amp) And phase weight (W)_s,phase)：

And

wherein the total number of parameters (amplitude and phase) of the whole combined MIMO DPD model for the s-th group of antenna branches is:

the # parameter is 2((1+ SK (M +1)) × S).

Note that W_s,ampAre labeled as:

·w_s,1,lwhich is the amplitude offset term for the l antenna branch in the s-th group of antenna branches,

·w_s,1+l(M+1),lwhich is the amplitude response directly input to the antenna branch i, and

·W_s,ampis the magnitude response to the current and past values of the different antenna branch inputs in the s-th group of antenna branches.

Likewise, W_s,phaseAre labeled as:

·w_s,1,lwhich is the phase offset term for the l antenna branch in the s-th group of antenna branches,

·w_s,1+l(M+1),lwhich is the phase response directly input to the antenna branch i, and

·W_s,ampis the phase response to the current and past values of the different antenna branch inputs in the s-th group of antenna branches.

Although not shown in the flow chart of fig. 14, the combined MIMO DPD system 1206-s for the s-th group of antenna branches is then applied to the input signal x_s,1To x_s,SApplying the calculated weight W to the samples of each of (1) when pre-distorting_s,ampAnd W_s,phaseTo provide a predistorted input signal u for the s-th group of antenna branches_s,1To u_s,SWhich is then processed by the respective antenna branch to output a transmit signal y for the s-th group of antenna branches_s,1To y_s,S。

Note that the process of FIG. 14 finds U_sAnd thus give the results for the s group dayThe PA 1202- (S, 1) to 1202- (S, S) of the line branch linearize the corresponding desired predistorted input signal of the sample of the input signal. The computational cost of performing linearization at a per sample level is too high; in contrast, U_sIs modeled as a function of the input samples, which is referred to herein as "ILC combined DPD".

Note that for the kernel decay parameter γ_kK:1 → K, optimization, EM can be used to jointly optimize these attenuation parameters as well as the amplitude and phase weights. Gradient descent for updating gamma_kK:1 → K. Further details and further explanation regarding this problem may be found in [14]]Is found in (1).

The packet-combining MIMO DPD shown in fig. 12 and 14 and described above provides advantages over the conventional MIMO DPD scheme by providing a good compromise between performance, computational complexity, implementation complexity and scalability.

Fig. 15 shows some example antenna branch groupings for a dual-polarized antenna array having 64 antenna elements per polarization. Thus, there are 128 antenna elements in total. Fig. 15 shows different group sizes. Figures 16 and 17 for a simulation of a particular embodiment show performance and complexity results using different set sizes. These simulation results show that the group of eight antenna branches seems to be a good compromise considering computational complexity, implementation complexity, performance and scalability. More specifically, as shown in fig. 16, for a group size of 8, the performance of the packet-combined DPD approaches that of the per-branch MISO DPD in terms of ACLR and EVM. As the group size increases (e.g., to 32, 64, or higher), performance decreases. A group size of 8 (or 16) is a good compromise. Also, fig. 17 shows that the number of FLOPS increases as the group size increases. Also, a group size of 8 (or 16) is a good compromise.

Fig. 18 shows four different options for MIMO DPD, namely Single Input Single Output (SISO) DPD, MISO DPD, combined MIMO DPD and grouped combined MIMO DPD. SISO DPD is known in the art. The present disclosure provides DPD schemes for MISO DPD (see, e.g., fig. 1 and 4 and corresponding description above), combined MIMO DPD (see, e.g., fig. 5 to 11 and corresponding description above), and grouped combined MIMO DPD (see, e.g., fig. 12 to 17 and corresponding description above).

Any suitable steps, methods, features, functions or benefits disclosed herein may be performed by one or more functional units or modules of one or more virtual devices. Each virtual device may include a plurality of these functional units. These functional units may be implemented via processing circuitry, which may include one or more microprocessors or microcontrollers, as well as other digital hardware, which may include Digital Signal Processors (DSPs), dedicated digital logic, and the like. The processing circuitry may be configured to execute program code stored in memory, which may include one or more types of memory such as Read Only Memory (ROM), Random Access Memory (RAM), cache memory, flash memory devices, optical storage, and the like. Program code stored in the memory includes program instructions for executing one or more telecommunications and/or data communications protocols and instructions for performing one or more of the techniques described herein. In some implementations, the processing circuitry may be configured to cause the respective functional units to perform corresponding functions in accordance with one or more embodiments of the present disclosure.

While the processes in the figures may show a particular order of operations performed by certain embodiments of the disclosure, it should be understood that such order is exemplary (e.g., alternative embodiments may perform the operations in a different order, combine certain operations, overlap certain operations, etc.).

At least some of the following abbreviations may be used in the present disclosure. If there is an inconsistency between abbreviations, the above usage should be preferred. If listed multiple times below, the first list should be prioritized over any subsequent lists.

AAS advanced antenna System

ACLR adjacent channel leakage ratio

ADC analog-to-digital conversion

DAC digital-to-analog conversion

DI Dual input

DI-PA dual input power amplifier

DPD digital predistortion

DSP digital signal processor

Expectation maximization of EM

EVM error vector magnitude

FDD frequency division Duplex

FLOPS floating-point operation

GMP generalized memory polynomial

ILA Indirect learning architecture

ILC iterative learning control

MIMO multiple input multiple output

MISO multiple input Single output

PA Power Amplifier

RAM random access memory

RBF radial basis function

RF radio frequency

ROM read-only memory

SISO Single input Single output

TDD time division duplexing

Those skilled in the art will recognize improvements and modifications to the embodiments of the present disclosure. All such improvements and modifications are considered within the scope of the concepts disclosed herein.

Reference to the literature

[1] T.l. marzetta, "non-cooperative cellular radio with an unlimited number of base station antennas", IEEE conference radio communication, vol 9, No.11, p 3590-.

[2]Pozar, "microwave engineering", 4 th edition, section 9.4, page 475-: wiley, 2011 (d.m. potar, "Microwave Engineering," 4)^th ed.,Sec.9.4,pp.475-482,Hoboken,NJ,USA:Wiley,2011)。

[3] Saffar et al, "Behavioral modeling of MIMO nonlinear systems with multivariate polynomials, section III," IEEE Confuction microwave Theory, volume 59, phase 11, p. 2994-.

[4] Bassam et al, "cross digital predistorter for compensating for crosstalk and non-linearity in MIMO transmitters," section III, "IEEE convention Microwave Theory technology, volume 57, phase 5, page 1119-.

[5] Abdldhafiz et al, "high performance reduced complexity behavioral models and digital predistorters for MIMO systems with crosstalk, section III," IEEE conference communication, volume 64, phase 5, page 1996-.

[6] Amin et al, "Behavioral modeling and linearization of crosstalk and memory effects in RF MIMO transmitters, sections II, IV," IEEE society of microwave Theory, Vol.62, No.4, p.810-.

[7] Hausman et al, "Prediction of Nonlinear Distortion in broadband Active Antenna Arrays, sections III, IV," IEEE convention microwave Theory technology, volume 65, phase 11, page 4550 + 4563, year 2017 for 11 months (k. hausman et al, "Prediction of Nonlinear Distortion in wireless band Active array, sec.iii, IV," IEEE trans.microwave Theory tech, vol.65, No.11, pp.4550-4563, nov.2017).

[8] Eun et al, "novel Volterra predistorter based on indirect learning architecture, section III," IEEE issue Signal Processing, volume 45, phase 1, page 223-.

[9] Barradas et al, "Digital predistortion of RF PA for MIMO transmitter based on equivalent load, section I, II, III," seminars integrated nonlinear microwave millimeter Wave Circuits, pages 1-4, 4 months 2017 (f.m. barradas et al, "Digital prediction of RF PAs for MIMO transmitters based on the equivalent load, sec.i, II, III," Workshop integration. nonlinear micro w.millimeter-Wave Circuits, pp.1-4, April 2017).

[10] chani-Cahuana et al, "Iterative Learning Control of RF Power Amplifier Linearization, section II, III," IEEE symposium microwave Theory technology, volume 64, phase 9, page 2778-.

[11] Isaksson et al, "broadband dynamic modeling of power amplifiers using radial basis function neural networks," section III, "IEEE Association microwave Theory, Vol.53, No.11, p.3422-.

[12] Zentono et al, "use Orthogonal non-parametric Kernel Smoothing Estimator to find Structural Information About RF Power amplifier," section II, "IEEE conference vehicle Technology, volume 65, phase 5, pages 2883-.

[13] Morgan et al, "generalized memory polynomial model for digital predistortion of RF power amplifiers," section I, "IEEE conference Signal processing, Vol.54, No.10, p.3852 & 3860, month 2006 10 (D.Morgan et al," A generated memory polymial model for digital predistortion of RF power amplifiers, Sec.I, "IEEE trans.Signal Process, vol.54, No.10, pp.3852-3860, Oct.2006).

[14] M.hamid et al, "Non-parametric spectral mapping using adaptive radial basis functions, section III," 2017IEEE international conference acoustics, Speech and Signal Processing (ICASSP), page 3599 + 3603, los angeles New Orleans in USA,2017 (m.hamid et al, "Non-parametric spectra using adaptive radial basis functions, sec.iii," 2017IEEE int.conf.acoustics, spech and Signal Processing (ICASSP), pp.3599-3603, New orans, LA, USA, 2017).