Predistortion method and system for a non-linear device-under-test

Abstract

The present disclosure relates to a predistortion method and a predistortion system for a non-linear device-under-test, DUT. The predistortion method comprises the steps of: providing a reference input waveform to the DUT; deriving a predistorted waveform for the DUT based on the reference input waveform using an iterative direct digital predistortion technique; analyzing a relationship between the reference input waveform and the calculated predistorted waveform using a mathematical model; deriving a predistortion algorithm for the DUT based on said analysis; and applying said predistortion algorithm to an input signal and feeding the, thus, predistorted input signal to the DUT.

Claims

1. A predistortion method for a non-linear device-under-test, DUT, comprising: providing a reference input waveform to the DUT; deriving a predistorted waveform for the DUT based on the reference input waveform using an iterative direct digital predistortion technique; analyzing a relationship between the reference input waveform and the derived predistorted waveform using a mathematical model, wherein the mathematical model comprises a Hammerstein model, a Wiener model, or a Volterra series model; deriving a predistortion algorithm for the DUT based on said analysis; and applying said predistortion algorithm to an input signal and feeding the, thus, predistorted input signal to the DUT.

2. The method of claim 1, wherein the predistortion algorithm compensates non-linear distortions as well as memory effects of the DUT.

3. The method of claim 1, wherein the mathematical model comprises a memory polynomial model.

4. The method of claim 3, wherein a starting point for the Hammerstein model is chosen randomly, and, in case the randomly chosen starting point does not converge, a different starting point is used.

5. The method of claim 1, wherein the step of analyzing the relationship between the reference input waveform and the derived predistorted waveform using the mathematical model comprises calculating parameters of the mathematical model, wherein said parameters depend on a hardware configuration of the DUT.

6. The method of claim 1, wherein a peak power of the reference waveform is increased while performing the iterative direct digital predistortion.

7. A predistortion method for a non-linear device-under-test, DUT, comprising: providing a reference input waveform to the DUT; deriving a predistorted waveform for the DUT based on the reference input waveform using an iterative direct digital predistortion technique; analyzing a relationship between the reference input waveform and the derived predistorted waveform using a mathematical model, wherein the mathematical model comprises a Hammerstein model, a Wiener model, or a Volterra series model; deriving a predistortion algorithm based on said analysis; and applying said predistortion algorithm in a signal generator.

8. The method of claim 7, wherein the predistortion algorithm compensates non-linear distortions as well as memory effects of the DUT.

9. The method of claim 7, wherein the signal generator is configured to perform a real-time predistortion of an input signal for the DUT based on said predistortion algorithm.

10. The method of claim 7, wherein the mathematical model comprises a memory polynomial model.

11. The method of claim 7, wherein the step of analyzing the relationship between the reference input waveform and the derived predistorted waveform using the mathematical model comprises calculating parameters of the mathematical model, wherein said parameters depend on a hardware configuration of the DUT.

12. A predistortion system for a non-linear device-under-test, DUT, comprising: a signal source configured to generate a reference input waveform to the DUT; a signal analyzer configured to receive an output waveform of the DUT, a processing unit configured to derive a predistorted waveform based on the reference input waveform and the output waveform using an iterative direct digital predistortion technique; wherein the processing unit is configured to analyze a relationship between the reference input waveform and the derived predistorted waveform using a mathematical model, wherein the mathematical model comprises a Hammerstein model, a Wiener model, or a Volterra series model, wherein the processing unit is configured to derive a predistortion algorithm for the DUT based on said analysis; and a predistortion unit configured to apply said predistortion algorithm to an input signal for the DUT.

13. The system of claim 12, wherein the predistortion unit is configured to perform a real-time predistortion of the input signal based on said predistortion algorithm.

14. The system of claim 12, wherein the system further comprises: a signal generator configured to feed the, thus, predistorted input signal to the DUT.

15. The system of claim 12, wherein the predistortion algorithm compensates non-linear distortions as well as memory effects of the DUT.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) The above described aspects and implementation forms of the present disclosure will be explained in the following description of specific embodiments in relation to the enclosed drawings, in which:

(2) FIG. 1 shows a flow diagram of a predistortion method for a non-linear DUT according to an embodiment;

(3) FIG. 2 shows a flow diagram of a step of the predistortion method shown in FIG. 1 according to an embodiment;

(4) FIG. 3 shows a flow diagram of a predistortion method for a non-linear DUT according to an embodiment;

(5) FIG. 4 shows a schematic diagram of a system for characterizing a non-linear DUT according to an embodiment;

(6) FIG. 5 shows a schematic diagram of a system for performing a predistortion of a DUT input waveform according to an embodiment; and

(7) FIG. 6 shows a schematic diagram of a predistortion system for a non-linear DUT according to an embodiment.

DETAILED DESCRIPTIONS OF EMBODIMENTS

(8) FIG. 1 shows a flow diagram of a predistortion method 10 for a non-linear DUT according to an embodiment.

(9) The predistortion method 10 comprises the steps of: providing 11 a reference input waveform to the DUT; deriving 12 a predistorted waveform for the DUT based on the reference input waveform using an iterative direct digital predistortion technique; analyzing 13 a relationship between the reference input waveform and the derived predistorted waveform using a mathematical model; deriving 14 a predistortion algorithm for the DUT based on said analysis; and applying 15 said predistortion algorithm to an input signal and feeding the, thus, predistorted input signal to the DUT.

(10) The DUT can be an amplifier, such as a power amplifier. The DUT can be configured to amplify signals according to the 5G technology standard, e.g. signals with bandwidths of up to 100 MHz. For instance, the input signal can be a 5G uplink signal.

(11) In particular, the predistortion algorithm compensates non-linearities caused by the DUT and memory effects of the DUT. Thus, in the step of applying 15 the predistortion algorithm to the input signal and, subsequently, feeding the input signal to the DUT, a real-time memory predistortion of the input signal can be performed.

(12) The iterative direct digital predistortion (iterative direct DPD) technique typically compares the reference input waveform to an output waveform of the DUT on a sample-by-sample basis (iteratively) and modifies each sample individually in amplitude and phase to derive the predistorted waveform. The iterative direct DPD may converge after a number of iterations, e.g. 5-10 iterations. Thus, in step 12, the predistorted waveform for the DUT can be derived by the iterative direct DPD based on the reference input waveform and a distorted output waveform of the DUT.

(13) In particular, the iterative direct DPD technique is model-independent, i.e. it is not limited by any model parameters. The iterative approach allows the modelling of memory effects of the DUT without increasing the complexity, because the memory effects are present in the measured samples.

(14) During iterative direct DPD, a peak power of the reference waveform can be increased in order to acquire a more precise model and/or to enhance the dynamic range. In particular, by enhancing the peak power in a non-linear range of the DUT, the effect of gain compression (i.e., the output signal progressively deviating from a linear relation to the input signal) can be reduced.

(15) The mathematical model used in the step 13 can take into account the real-time hardware of the DUT when deriving the parameters that correlate the reference input waveform with the predistorted waveform. In other words, in the step 13 parameters of the mathematical models are derived, said parameters being based on the specific hardware of the DUT.

(16) For example, the mathematical model can comprise a Hammerstein model, a Wiener model, a Volterra series model, a memory polynomial model or another suitable polynomial model. In particular, these mathematical models are capable to characterize memory effects of the DUT.

(17) The Volterra series can be used to model time invariant non-linear dynamic systems and can, therefore, be applied to model the non-linearity and memory effects of a DUT, e.g. a power amplifier.

(18) The Volterra Series model can be written in a discrete form as:

(19) $\begin{array}{c}{y}_{\mathrm{VS}}\left[n\right]=\underset{m_1=0}{\overset{M}{.Math.}}{h}_1\left(m_1\right)x\left[n-m_1\right]\\ +\underset{m_1=0}{\overset{M}{.Math.}}\underset{m_2=0}{\overset{M}{.Math.}}{h}_2\left(m_1,m_2\right)x\left[n-m_1\right]x\left[n-m_2\right]\\ .Math.\\ +\underset{m_1=0}{\overset{M}{.Math.}}.Math.\underset{m_K=0}{\overset{M}{.Math.}}{h}_2\left(m_1,.Math.,m_2\right)x\left[n-m_1\right].Math.x\left[n-m_K\right],\end{array}$
Thereby, h.sub.n are Volterra kernels, M is a memory depth, K is an order of the non-linearity and x[n] is an input signal. The Volterra series can be described as a linear superposition of non-linear effects. Therefore, it is possible to calculate the kernels (parameters of the model) linearly.

(20) The memory polynomial model can be obtained from the Volterra series, in particular when only considering samples of the input vector with the same time shift and neglecting the cross terms. By not using the cross terms of the Volterra series, the memory polynomial model reduces the number of parameters, thus, offering a higher performance and lower complexity as compared to the Volterra series.

(21) The Hammerstein model is a box-oriented model with a reduced complexity compared to the memory polynomial model. The Hammerstein model consists of a filter and a non-linearity. Thereby, the non-linearity is static with all time-dependent components being compensated with the filter, e.g. a Finite Impulse Response (FIR) filter. The Wiener model is similar to the Hammerstein model with the difference between these models being a reverse order of filter and non-linearity.

(22) An output signal of the Hammerstein model can be written as follows:

(23) $y_{\mathrm{HM}}\left[n\right]=\underset{m=0}{\overset{M-1}{.Math.}}\underset{\stackrel{\bigvee }{k}=1}{\overset{K}{.Math.}}{h}_m{c}_{k}x\left[n-m\right]{.Math.x\left[n-m\right].Math.}^{k-1},$
with M being the memory length, K being the polynomial degree, c.sub.k being complex coefficients of the non-linearity and h.sub.m being complex coefficients of the FIR filter. The coefficient c.sub.k and h.sub.m can be stacked into vecotrs c and h. The Hammerstein Model can be written as a memory polynomial model with a memory polynomial model parameter matrix A=c.Math.h.sup.T. The memory polynomial coefficients (filter and non-linearity) can be calculated based on a Least Squares Problem approach, for example, using a normalized iterative algorithm or other suitable approach. In particular, the non-linearity is thereby described by the elements of c and the filter by the elements of h.

(24) To derive the parameters filter and non-linearity of the Hammerstein model with this approach, a starting point for either c or h is first chosen. This starting point can be chosen randomly. With a good starting point, the algorithm usually converges in less than 50 iterations. Thus, the parameters can be determined by choosing a starting point, let the algorithm compute c and h and if it did not converge after a number of iterations, e.g. 50 iterations, use a different starting point.

(25) The step 13 of analyzing the relationship may comprise using the Hammerstein model to fit the reference waveform to the predistorted waveform, which was derived via direct iterative DPD (step 12), and determining parameters (e.g., the coefficients for filter and non-linearity) of the fitted Hammerstein model based on said fitting. In particular, the determined parameters of the Hammerstein model depend on the specific hardware configuration of the DUT and take into account non-linearities and/or memory effects of the DUT.

(26) FIG. 2 shows a flow diagram of sub-steps of step 13 of the predistortion method 10 shown in FIG. 1 according to an embodiment. According to the above, the step 13 comprises the sub-step 16 of choosing a starting point for the Hammerstein model, e.g. choosing the starting point randomly, and deriving parameters of the Hammerstein model, e.g. based on a fitting of the Hammerstein model to the predistorted waveform derived via iterative direct DPD in step 12. If the chosen starting point does not converge after a number of iterations, a different starting point can be chosen.

(27) The predistortion algorithm can then be derived (step 14) based on the determined coefficients from the Hammerstein model.

(28) Based on the determined parameters of the Hammerstein model (filter and non-linearity) an input signal for the DUT can be modified, e.g. by adapting a signal generator, which generates and the input signal, according to these parameters. The, thus, predistorted input signal can then be fed to the DUT.

(29) FIG. 3 shows a flow diagram of another predistortion method 30 for the non-linear DUT according to an embodiment.

(30) The predistortion method 30 comprises the steps of: providing 31 a reference input waveform to the DUT; deriving 32 a predistorted waveform for the DUT based on the reference input waveform using an iterative direct digital predistortion technique; analyzing 33 the relationship between the reference input waveform and the derived predistorted waveform using a mathematical model; deriving 34 a predistortion algorithm based on said analysis; and applying 35 said predistortion algorithm in a signal generator.

(31) The method 30 shown in FIG. 3 can be performed in a test and measurement instrument or setup for any given DUT. The DUT can be a non-linear amplifier, such as a power amplifier.

(32) In particular, the method steps 31 to 34 of the method 30 shown in FIG. 3 can be identical to the method steps 11 to 14 of the method 10 shown in FIG. 1.

(33) For example, the signal generator has a configurable non-linearity and a configurable filter designed into its hardware. The non-linearity is, for example, user adjustable. The non-linearity and filter of the signal generator can be represented by a Hammerstein model. Thus, the step of applying 35 the predistortion algorithm in a signal generator, may comprise applying the predistortion algorithm derived in step 34 (which contains the parameters of the Hammerstein Model) to the signal generator and, thereby, adjusting the configurable non-linearity and filter of the signal generator based on the derived non-linearity and filter according to the Hammerstein model.

(34) In particular, the signal generator is configured to perform a real-time predistortion of an input signal for the DUT based on the applied predistortion algorithm. The signal generator can comprise a predistortion unit which is configured to apply said predistortion algorithm to an input signal for the DUT. The predistortion algorithm can take into account non-linear distortions as well as memory effects of the DUT.

(35) FIG. 4 shows a schematic diagram of a system 40 for characterizing a non-linear DUT 43 according to an embodiment.

(36) The system 40 can be configured to characterize the DUT 43 using iterative direct DPD (MC, iterative learning control), ideally at the operating point of DUT 43. Thereby, a signal generator 42 provides the DUT 43 with an input waveform 41. A signal analyzer 44, e.g. a spectrum analyzer, an oscilloscope or a network analyzer, provides a distorted DUT output waveform from the DUT 43. An iterative direct DPD algorithm using the DUT input and output waveform can be used to calculate a predistorted DUT input waveform. This direct DPD algorithm, for instance, uses a gain expansion technique to maximize the characterization range regarding a DUT input level. Subsequently, a Hammerstein model 45 can be derived by fitting a DUT input waveform to the predistorted DUT input waveform.

(37) FIG. 5 shows a schematic diagram of a system 50 for performing a predistortion of a DUT input waveform according to an embodiment.

(38) The system 50 can be configured to perform a real-time predistortion of a DUT input waveform 51. Therefore, a digital predistortion unit 56 of the system uses the derived Hammerstein model 55 (e.g., derived with the system 40 in FIG. 4) to generate a predistorted DUT input waveform comprising a memory-less nonlinearity and a linear filter. A signal generator 52 can provide the, thus, predistorted input waveform to the DUT 53. The predistortion unit 56 can be a component of the signal generator 52. Optionally, a signal analyzer 54, e.g. a spectrum analyzer, an oscilloscope or a network analyzer, can provide measurement results to verify a DPD performance.

(39) FIG. 6 shows a schematic diagram of a predistortion system 60 for the non-linear DUT 63 according to an embodiment.

(40) The system 60 comprises a signal source 61 configured to generate a reference input waveform 62 to the DUT 63, a signal analyzer 66 configured to receive an output waveform 65 of the DUT 63, and a processing unit 67 configured to derive a predistorted waveform based on the reference input waveform 62 and the output waveform 65 using an iterative direct digital predistortion technique. The processing unit 67 can be configured to analyze a relationship between the reference input waveform 62 and the derived predistorted waveform using a mathematical model, and to derive a predistortion algorithm for the DUT 63 based on said analysis. The system 60 further comprises a predistortion unit 68 which is configured to apply said predistortion algorithm to an input signal 69 for the DUT 63.

(41) The system 60 can form a test environment for the DUT 63. The system 60 can be an implementation of a real-time DPD that can be derived by using the methods 10, 30 shown in FIGS. 1 to 3. In particular, the system 60 can be configured to carry out any one of the methods 10, 30 shown in FIGS. 1 to 3.

(42) The DUT 63 can be an amplifier, such as a power amplifier. The DUT 63 can be configured to amplify signals according to the 5G technology standard, e.g. signals with bandwidths of up to 100 MHz. For instance, the input signal is a 5G uplink signal.

(43) In particular, the predistortion algorithm compensates non-linearities and/or memory effects of the DUT 63. Thus, in the step of applying 15 the predistortion algorithm to the input signal and, subsequently, feeding the input signal to the DUT 63, a real-time memory predistortion of the input signal can be performed.

(44) The signal analyzer 66 can be a spectrum analyzer, an oscilloscope or a network analyzer. The signal analyzer 66 can provides measurement results to verify a DPD performance. The processing unit 67 can be a component of the signal analyzer 46.

(45) The predistortion unit 68 can be configured to perform a real time predistortion of the input signal 69 based on the predistortion algorithm. For example, the predistortion unit 68 can comprise a configurable non-linearity and filter. The predistortion unit 68 can be configured to adjust said configurable non-linearity and filter based on parameters of the derived predistortion algorithm, e.g. parameter coefficients of a Hammerstein model.

(46) The system 60 can further comprise a signal generator. The signal generator can be configured to feed the, thus predistorted input signal 69 to the DUT 63. For example, the predistortion unit 68 can be a component of the signal generator or can be connected to the signal generator. The signal generator can the identical to the signal source 61 that was used to generate the reference input waveform.

(47) In particular, the system 40 shown in FIG. 4 may be a part the system 60 shown in FIG. 6. For instance, the signal generator 42 and the signal analyzer 44 of the system 40 in FIG. 4 may correspond to the signal source 61 and the signal analyzer 66 of the system 60 in FIG. 6. Also the DUT 43 may be identical to the non-linear DUT 63 in FIG. 6. In particular, the system 40 in FIG. 4 may correspond to the part of the system 60 in FIG. 6 that is configured to characterize the DUT 43, 63 in order to derive the mathematical model.

(48) Likewise, the system 50 shown in FIG. 5 may be a part the system 60 shown in FIG. 6. For instance, the signal generator 52, the signal analyzer 54 and the predistortion unit 56 of the system 50 in FIG. 5 may correspond to the signal source 61, the signal analyzer 66 and the predistortion unit 68 of the system 60 in FIG. 6. Also, the DUT 53 may be identical to the non-linear DUT 63 in FIG. 6. In particular, the system 50 in FIG. 5 may correspond to the part of the system 60 that is configured to apply the predistortion algorithm to an input signal of the DUT 53, 63.

(49) All features of all embodiments described, shown and/or claimed herein can be combined with each other.