Circuit and method for compensating for nonlinearity

Abstract

A circuit (100) for compensating for nonlinearity without essentially changing an operating point of a characteristic curve and / or an operating range of a characteristic curve is described, the circuit (100) comprising an alternating voltage signal source (10) for providing an input signal (20), a control unit (30) for receiving the input signal (20) and converting the input signal (20) into a predistorted signal (40) according to at least one predefined predistortion parameter, and a sink (50) for receiving the predistorted signal (40), the sink (50) being adapted to provide an adjustment signal to the sink (50) for operating the sink (50) in an operating range or at an operating point. and the control unit (30) is configured to receive at least one sensor signal (70) of the sink (50) in a feedback manner and to adapt at least one preset pre-distortion parameter based on the at least one sensor signal (70), the control unit (30) converting the input signal (20) into a pre-distorted signal (40) by the at least one adapted pre-distortion parameter in order to provide the pre-distorted signal (40) to the sink (50) without essentially changing an operating point of the characteristic curve and / or an operating range of the characteristic curve. Additionally, a method for compensating for nonlinearity without essentially changing an operating point of the characteristic curve and / or an operating range of the characteristic curve is described.

Description

[Technical Field]

[0001] The present invention relates to a circuit for compensating for nonlinearities without essentially changing the operating point of a particular curve and / or the operating range of a characteristic curve, and to a circuit for compensating for nonlinearities without essentially changing the operating point of a characteristic curve and / or the operating range of a characteristic curve. [Background technology]

[0002] In operation, system nonlinearities result in undesired deviations in system behavior from a purely linear relationship between input signals and system response. These nonlinearities can be compensated for, especially using inverse functions, after identifying the nonlinearities, for example, as characteristic curves. However, this is particularly problematic when the system is operating at an operating point that is not intended to change (e.g., using a DC bias) and / or when operating within an operating range (e.g., using a minimum or maximum input voltage swing). Using an inverse function to compensate for system nonlinearities can also result in undesired changes in system output levels.

[0003] In TUMPOLD, David et al., "Linearizing an electrostatically driven MEMS speaker by applying pre-distortion," Sensors and Actuators A: Physical, 2015, 236th edition, pp. 289-298, a pre-distortion function for an electrostatic MEMS loudspeaker is discovered using a local model network and direct inverse control. The method described in TUMPOLD is more complex than the novel method described here, both in terms of implementation and collection of the required data. The operating point and operating range of the MEMS loudspeaker are not automatically maintained by this pre-distortion function.

[0004] In the disclosure of MOORE, Steven Ian et al., "Feedback-Controlled MEMS Force Sensor for Characterization of Microcantilevers," Journal of Microelectromechanical Systems, 2015, 24th edition, No. 4, pp. 1092-110, predistortion of an electrostatic sensor is implemented as an analog circuit using a square root function. The operating range and operating point are not particularly considered, and are not automatically maintained by this predistortion function. Only second-order distortion can be compensated.

[0005] MOSCA, Simona, "Improving the virgo detector sensitivity: Effect of high power input beam and electrostatic actuators for mirror control," Doktorarbeit, Universita degli Studi di Napoli Federico II, 2009, p. 78, describes a method for predistorting the control signal of an electrostatic actuator. Here, a square-root function is amplitude-modulated with a correspondingly high modulation frequency so that approximately only the DC and linear portions remain. The operating point and operating range of the actuator are not automatically maintained by this predistortion function and may have to be entered manually. Only second-order distortion can be compensated.

[0006] German Patent No. 382177C describes the derivation of an inverse function for reducing harmonics or generating desired harmonics in high-frequency technology. This inverse function is derived directly from the system characteristic curve, for example by geometric construction, which must be known, or from the relationship between input and output amplitudes. Maintaining the operating point and operating range of the system does not occur automatically.

[0007] DE 3307309 C2 describes a method for transmitting electrical signals, in which the signal to be transmitted is predistorted before being fed to the transmitting element, and the operating point and operating range are not taken into account in this procedure at all. Maintaining the operating point and operating range of the system does not occur automatically. The system described in DE 3307309 C2 is significantly different from the invention described here. There are special requirements for the input signal, for example the absence of certain frequency components, and in addition the predistortion is performed by a polynomial.

[0008] US Patent No. 4,618,808A discloses compensation of second order distortion using a square root function, where the operating range and operating point are not specifically considered and are not automatically maintained by the predistortion function.

[0009] US Patent No. 6,597,650 B2 discloses a parameterized hyperbolic function for compensating, in particular, second-order distortion in a transmission system / data carrier reading system, the operating range of the system being only roughly maintained. [Prior art documents] [Patent documents]

[0010] [Patent Document 1] German Patent No. 382177 [Patent Document 2] German Patent No. 3307309 [Patent Document 3] U.S. Patent No. 4,618,808 [Patent Document 4] U.S. Patent No. 6,597,650 [Non-patent literature]

[0011] [Non-Patent Document 1] TUMPOLD, David et al., Linearizing an electrostatically driven MEMS speaker by applying pre-distortion, Sensors and Actuators A: Physical, 2015, 236th edition, pp. 289-298 [Non-patent document 2] MOORE, Steven Ian et al., Feedback-Controlled MEMS Force Sensor for Characterization of Microcantilevers, Journal of Microelectromechanical Systems, 2015, 24th edition, No. 4, pp. 1092-1101 [Non-patent document 3] MOSCA, Simona, Improving the virgo detector sensitivity: Effect of high power input beam and Electrostatic actuators for mirror control, Doktorarbeit, Universita degli Studi di Napoli Federico II, 2009, page 78 Summary of the Invention [Problem to be solved by the invention]

[0012] DC correction frees the equalized signal from its mean value in some implementations. The measure of nonlinear distortion only relates to second-order distortion of pure or sinusoidal tones.

[0013] The objective behind the present invention is to provide a circuit and method whereby the operating point and / or operating range can be automatically maintained while non-linear distortion at the output of the circuit can be reduced by pre-distorting the signal. [Means for solving the problem]

[0014] This object is achieved by the subject matter of the independent claims.

[0015] The core idea behind the present invention is to provide a circuit that can implement a method, particularly during operation, that allows a sink to reduce nonlinear predistortion of the signal provided to it, while automatically maintaining the operating point and / or operating range, particularly through predistortion established during operation of the circuit. In other words, using the proposed method and / or circuit, a signal that is nonlinear compared to the original input signal can be transferred to the sink. The signal ultimately transferred to the sink is predistorted so that the signal at the output of the sink, referred to herein as the sensor signal, is as linear as possible compared to the original input signal. Additionally, the sink can be operated at a constant operating point and / or operating range. The predistortion is performed so that the operating point does not change, i.e., there is no DC component due to the predistortion, and the operating range is maintained, i.e., the original AC input voltage range is not exceeded, but only slightly below it as much as possible. Specifically, no special input signals, particularly input signals that deviate from conventional input signals, are required, and no system characterization needs to be performed before operation. Rather, conventional signals can be used.

[0016] The term "as linear as possible" should be understood as follows. First, there is the input signal. The input signal is linear, or more specifically, identical, compared to itself. The predistortion performs various linear and nonlinear transformations, resulting in a predistorted signal that is somewhat more nonlinear than the input signal. The predistorted signal is finally transferred to a sink, which is a nonlinear system. At the output of the sink, monitored by sensor technology, one or more sensor signals are generated that are nonlinear compared to the predistorted signal. When the predistortion is optimally adjusted by the proposed circuit and method, in the best case, the sensor signal at the output of the sink is linear compared to the input signal. However, depending on the characteristics of the predistortion function and the sink used, the optimum condition may be a sensor signal that is only approximately linear with respect to the input signal.

[0017] The proposed circuit for compensating for nonlinearity without substantially changing the operating point and / or operating range of a characteristic curve comprises an AC voltage signal source for providing an input signal, a control unit for receiving the input signal and converting the input signal into a predistorted signal according to at least one preset predistortion parameter, and a sink for receiving the predistorted signal, the sink being coupled to the adjustment unit configured to provide an adjustment signal to the sink to operate the sink within an operating range or at an operating point. The control unit is configured to receive at least one sensor signal of the sink in a feedback manner and adapt the at least one preset predistortion parameter based on the at least one sensor signal, which may also be referred to as a sink output signal. The preset parameter can be adjusted so that the predistortion initially has little effect on the signal. Exemplarily, the predistortion can be implemented so that it is inaudible or causes interfering movements of a low amplitude. The predistorted signal is transmitted to the sink, which reacts to it, resulting in a certain system response of the sink. This system response, i.e., the sink output signal, is then provided at the output of the sink. The feedback scheme here should be understood as one in which the sink output signal, also called the sensor signal in this case, or a measurement based thereon, is fed back to the control unit. The sensor signal is then applied back to an input of the control unit, which is then applied to an input of the control unit that is different from the original input signal. This is the feedback path of the control. In this case, the terms control unit, control device, control means, controller, etc. are used synonymously.

[0018] The control unit in this feedback path operates as follows: The original AC input signal is predistorted by the control unit. The predistortion is adjusted using one or more parameters. These parameters have initial values ​​(e.g., so that the predistortion initially has minimal effect). The predistorted signal is then transmitted to a sink that reacts to the signal, and a system response, in this case referred to as a sensor signal, is measured. The sensor signal is fed back to a separate input of the control unit. Based on the sensor signal or a derived measurement of the nonlinearity of the sensor signal, the predistortion parameters of the control unit are then adapted so that predistortion of the original input signal ultimately results in minimal predistortion at the output of the sink. The signal predistorted by the new set of parameters is then transmitted back to the sink, and so on. This describes a control loop with a feedback path. The control unit transforms the input signal into a predistorted signal by at least one adapted predistortion parameter in order to provide the predistorted signal to the sink without essentially changing the operating point and / or operating range of the characteristic curve.

[0019] "Without essentially changing the operating point and / or operating range of the characteristic curve" can be understood here as maintaining the operating point, especially the DC offset, and changing the operating range "as little as possible," with the important boundary condition that the maximum value of the original operating range must never be exceeded. This means that the original operating range is maintained, but only a portion of it is used, and one of the extreme values ​​may always be pushed to the limit. This boundary condition becomes unclear when optional level compensation is used, which may partially exceed the original operating range. Alternatively, the operating range can be maintained when the DC portion generated by the predistortion function is adapted accordingly rather than completely removed. Thus, a correspondingly selected DC portion can allow the operating range of the input signal to be maintained. This therefore introduces an additional DC portion, which then changes the operating point of the sink. In addition, there is a further alternative in which both the operating point and the operating range remain unchanged. Alternatively, a balancing of both amounts of deviation between the predistorted signal and the input signal can be performed. In this case, both the operating point and the operating range deviate "a little" from the input signal.

[0020] The following example further explains the term "without essentially changing the operating point of the characteristic curve and / or the operating range of the characteristic curve."

[0021] Sink y=(1+x) 3 is equalized in such a way that the operating point (DC offset = 1) is maintained. Therefore, of the original operating range [-1:1] (i.e., for example, when x=sin(2*π*f*t)), only the range [-1:0.4] is used in the end. The operating range of the predistorted signal is within the original range and pushes the signal up to its limit, which is the extreme value (-1). The peak-to-peak signal stroke in this example is still 70% of the original signal. This example represents the so-called negative extreme case. Usually, the operating range of the predistorted signal is close to the operating range of the input signal, e.g. [-1:0.95].

[0022] The control loop described immediately above can reduce distortion in the sensor signal. Predistortion parameters are changed according to the control loop, and the predistorted signal is transmitted to a sink that reacts to the signal, resulting in a sensor signal with a certain distortion and therefore a certain value of the target function. The predistortion parameters are adapted so that the sensor signal becomes less distorted over time. The target function is discussed below.

[0023] The terms DC portion and DC offset are used synonymously herein.

[0024] The AC voltage signal source here may provide a digital or analog signal. Signal processing using the proposed circuit can be performed using a digital or analog signal. In particular, an AC voltage signal here should be understood as being discrete amplitude values ​​at regular sampling times, which arrive at the control unit. However, it is also possible for the control unit to be provided with an analog signal, which can be converted into a digital signal by the control unit. It is also conceivable that the control unit can convert the digital signal back into an analog signal before the signal is transferred to the sink. The algorithms realizing the methods described herein by the described circuits can be implemented for analog or digital signals.

[0025] The predistorted signal here may be a signal with a DC portion, without a DC portion, and / or with a known range of values, the range of values ​​having maximum and minimum values ​​that the predistorted signal can take.

[0026] Preferably, the control unit is configured to adapt or remove the DC portion when the predistorted signal includes the DC portion, specifically, to calculate the DC portion using the predistortion parameters and subtract the calculated DC portion from the predistorted signal when removing the DC portion, and / or to remove the DC portion from the predistorted signal by a high-pass filter with a sufficiently low cutoff frequency. Here, the calculated DC portion is first subtracted from the distorted signal, and then the predistorted signal from which the calculated DC portion is subtracted is sent through a high-pass filter when an "and" operation is realized. Here, "sufficient" means "maintaining the bandwidth of the original signal." A high-pass filter with a sufficiently low fundamental frequency is a high-pass filter that provides, after filtering, a filtered signal with the bandwidth of the original signal. In these embodiments, the operating point may be maintained. However, embodiments in which the DC portion is adapted so that the operating point is maintained are also conceivable. In addition, paths between these extremes are possible and also require adaptation of the DC portion. Here, the predistorted signal applied directly to the sink should not be confused with the DC portion that may be included in the predistorted signal. The DC portion of the predistorted signal is not a separate signal, but a part of the predistorted signal. This DC portion is applied to the sink only if it is not completely removed during predistortion. Specifically, if the operating point is to be maintained, the DC portion is removed. However, if the operating range is to be maintained, the predistorted signal at the input of the sink will include a DC portion that is adapted accordingly.

[0027] Using the proposed circuit, non-linear distortions in the sensor signal are minimized over time, in particular the proposed circuit comprises a control loop such that non-linear distortions in the sensor signal are reduced, in particular such that they disappear, especially during operation of the circuit.

[0028] A further aspect of the present invention relates to a method for compensating for nonlinearity without substantially changing the operating point and / or operating range of a characteristic curve, the method comprising the steps of providing an input signal by an AC voltage signal source, receiving the input signal by a control unit, and converting the input signal into a predistorted signal according to at least one preset predistortion parameter. The method then comprises receiving the predistorted signal by a sink, the sink being coupled to an adjustment unit. Simultaneously with receiving the predistorted signal by the sink, the method comprises providing an adjustment signal by the adjustment unit to the sink for operating the sink within an operating range and / or an operating point. The proposed method then comprises receiving at least one sensor signal output by the control unit to the sink in a feedback manner for adapting the at least one preset predistortion parameter based on the at least one sensor signal. Converting the input signal into a predistorted signal by the at least one adapted predistortion parameter is then performed to provide the predistorted signal to the sink without substantially changing the operating point and / or operating range of the characteristic curve. The proposed method reduces distortions in the sink output, i.e., the sensor signal. The predistorted signal is modified by stepwise / continuous parameter adaptation so that nonlinear distortions in the sensor signal are minimized. The explanations of the terms used are also valid when using these terms in the described method. It should be understood that the method can be performed by the proposed circuit, or a circuit can be configured to realize the method.

[0029] Advantageous implementations of the invention are the subject of the dependent claims. Preferred embodiments of the invention are discussed in more detail below with reference to the accompanying drawings. [Brief explanation of the drawings]

[0030] [Figure 1] 1 is a general signal flow diagram of the proposed circuit. [Figure 2] 4 is a signal flow chart of the proposed circuit according to a first variant. [Figure 3] 3 is a more detailed signal flow chart according to FIG. 2; [Figure 4] 4 is a more detailed signal flow chart according to FIGS. 2 and 3, divided into a first part and a second part for the sake of readability; [Figure 5] FIG. 10 shows simulation results without level compensation for the first variant of the method. [Figure 6] FIG. 10 shows simulation results including level compensation for a first variant of the method. [Figure 7] FIG. 10 shows simulation results including level compensation for a first variant of the method. [Figure 8] FIG. 10 shows simulation results including level compensation for a first variant of the method. [Figure 9a] 4 is a signal flow chart of the proposed circuit according to a second variant. [Figure 9b] 4 is a signal flow chart of the proposed circuit according to a second variant. [Figure 10a] 10 is a more detailed signal flow chart according to FIG. 9; [Figure 10b] 10 is a more detailed signal flow chart according to FIG. 9; [Figure 11a] FIG. 10 shows simulation results for a second variant of the method. [Figure 11b] FIG. 10 shows simulation results for a second variant of the method. [Figure 12] FIG. 10 shows simulation results for a second variant of the method. [Figure 13] FIG. 10 is a diagram showing simulation results of adaptive distortion when changing the operating point. [Figure 14] 1 is a flowchart of the proposed method. DETAILED DESCRIPTION OF THE INVENTION

[0031] Individual aspects of the present invention described herein are illustrated below in Figures 1 to 14. Figures 1 to 14 together explain the principles of the present invention. In this application, like reference numbers refer to like elements or elements of like effect, and when repeated, all reference numbers will not be discussed again in all figures.

[0032] All explanations of terms provided in this application can be applied to both the proposed circuit and the proposed method. To avoid redundancy wherever possible, explanations of terms will not be repeated multiple times.

[0033] FIG. 1 shows the components of a circuit 100, and therefore of a general method 140. An AC signal source 10 or AC voltage source 10 outputs an input signal x, also referred to by the reference number 20. The absolute minimum and maximum values ​​that the input signal x can take are known. It is not possible for the input signal to fall below the absolute minimum or exceed the absolute maximum. This is a prerequisite for maintaining the operating range. The input signal 20 is predistorted by a control unit 30. The control unit 30 therefore outputs a predistorted signal 40 at its output. The control unit 30 is configured to perform predistortion on the input signal 20. As a result, the predistorted signal 40 output by the control unit 30 can also be understood as being a predistorted output signal. The predistorted signal 40 is a predistorted AC signal

number

number

[0034] The control unit 30 requires a transmission path for an analog voltage or a digital signal, which results in a predistorted signal and is transmitted to the sink. The question of analog or digital depends on the configuration of the control unit 30 and the sink 50 (analog I / O, digital I / O). The transmission of the signal can basically be done in different ways. Preferably, electrical signals are handled, which is why the transmission path is usually a cable. An optical transmission path or other transmission paths known to those skilled in the art are also conceivable.

[0035] An operating point that changes only slowly over time depending on the application means that the operating point changes slowly enough that the optimization of the predistortion parameters, which is performed in parallel with normal operation, can converge (at least approximately). The speed of convergence depends mainly on the optimization hyperparameters, the nonlinear characteristic curve of the sink, and the statistical properties of the input signal. In the case of a changing operating point, the speed of convergence also depends on how strongly / by how much the operating point is moved. Too many strong changes, one after the other, too quickly will eventually prevent convergence and therefore proper predistortion.

[0036] Using the audio example, a specific implementation can be used to derive rough values ​​for the rate and magnitude of change. To guarantee convergence for most cases, the rate of change of the operating point must be very high, specifically, at least three orders of magnitude below the lowest frequency of the input signal. The faster / more frequently the operating point changes, the smaller the magnitude of the operating point change must be, specifically, the operating point should not change by more than a factor of three per minute. Good examples of operating point changes are slow, continuous material fatigue / degradation or slow heating of a device.

[0037] The sink's output signal 70 or several output signals 70 are referred to herein as sensor signals 70. The sensor signals 70 are fed back to the control unit 30, which uses these data either directly (as shown in Figures 3 and 4, for example, showing a first variant) or indirectly (as shown in Figures 9 and 10, for example, showing a second variant) to adapt the predistortion parameters.

[0038] The sensor signal 70 can be, for example, a measured output voltage, current intensity, sound pressure, or surface vibration. The sensor signal 70 can essentially be any physical quantity detectable by measurement technology. It can also be a measured temperature, possibly static, with a nonlinear correlation to the input signal 20. Specifically, the sensor signal 70 includes a measurement value, which is not itself a physical quantity, except in some cases for analog / digital voltage signals at the output of the sink, since these are compatible with the input of the control unit. Surface vibrations as physical quantities cannot be input to the control unit, only measurements detected by the corresponding sensor and, if applicable, processed by it. This fact should be clear to those skilled in the art and will not be discussed further. If the sink is purely digital / virtual, it can simply be a numerical value without a physical equivalent.

[0039] The proposed circuit 100 for compensating for nonlinearity without essentially changing the operating point and / or operating range of a characteristic curve comprises an AC voltage signal source 10 for providing an input signal 20, a control unit 30 for receiving the input signal 20 and converting it into a predistorted signal 20 by at least one preset predistortion parameter r, and a sink 50 for receiving the predistorted signal 40, the sink 50 being coupled to an adjustment unit 60 configured to provide an adjustment signal to the sink 50 to operate the sink 50 in an operating range or at an operating point. The control unit 30 is configured to receive at least one sensor signal 70 of the sink 50 in a feedback manner and to adapt the at least one preset predistortion parameter r based on the at least one sensor signal 70, which may also be referred to as a sink output signal 70. It is conceivable that the at least one predistortion parameter r is a real-valued variable. However, predistortion functions with several parameters are also conceivable. In this case, r can be a vector r, or else the parameters can be r1, r2, ..., r for M parameters. M must be indexed to read the vector r. m where m=1, 2, …, M, and M is a natural number.

[0040] The control means 30 converts the input signal 20 into a predistorted signal 40 by at least one adapted predistortion parameter in order to provide the predistorted signal 40 to the sink 50 without essentially changing the operating point of the characteristic curve and / or the operating range of the characteristic curve. With the described control loop, the input signal 20 provided by the AC voltage source 10 is applied to the input of the control unit 30 so that the input signal 20 can be predistorted with the at least one adapted distortion parameter. Here, the sensor signal 70 is fed back in the control loop so that the predistortion parameters can be adapted. However, the input signal 20 is always input to the control unit 30. The adjustment of the at least one distortion parameter r is repeated in the loop, in particular in parallel with the normal operation of the sink.

[0041] The predistorted signal 40 can be with or without a DC portion and / or within a known range of values, where the range of values ​​has maximum and minimum values ​​that the predistorted signal can have. If the predistorted signal is within the range of values, the operating range can be maintained. Since the minimum and maximum values ​​of the input signal are usually known, it is absolutely guaranteed that the predistorted signal 40 can be kept within this range of values.

[0042] If the predistorted signal 40 contains a DC offset, the control unit 30 is configured to modify or remove the DC offset, in particular to calculate the DC offset by an average value using at least one predistortion parameter r and subsequently subtract the calculated DC offset from the predistorted signal 40, and / or to remove the DC offset from the predistorted signal by a high-pass filter with a sufficiently low cut-off frequency. A high-pass filter with a sufficiently low cut-off frequency means that the bandwidth of the input signal 20 is maintained when passing through the high-pass filter. The bandwidth of the distorted signal after the high-pass filter corresponds to the bandwidth of the original signal, i.e., the input signal 20.

[0043] The predistorted signal 40 may contain a DC offset / DC portion, for example, if the operating range is to be maintained. However, if the operating point is to be kept unchanged, the DC offset may not be included, as it can be summed with the DC signal of the adjustment unit 60. It should be mentioned that depending on how the sink is implemented, the addition of the DC offset of the predistortion and adjustment unit does not necessarily occur. The DC offset is usually caused by the predistortion function. Depending on the goal (operating point or operating range kept unchanged), it is removed or not. It is also conceivable to pursue both goals (operating point and operating range kept unchanged) so as not to allow complete removal of the DC portion.

[0044] FIG. 2 discloses the mode of functioning of the control unit 30 in a first variant for operating the proposed circuit according to the proposed method. The first variation follows a feedback-based control paradigm. The predistortion parameters r / r are changed to minimize a first target function 80. Then, parameter variations 90 are performed. The first target function 80 is calculated based at least on one or more sensor signals 70, or additionally using a known input signal 20. Specifically, in some methods, the error function 80 always maps the nonlinearity of the sensor signal 70 to the input signal 20. Different measurement / target functions can be selected for this purpose. Two examples are described below.

[0045] First Example: If the input signal 20 consists solely of sinusoidal tones with known frequencies, the total harmonic distortion, distortion factor, and intermodulation distortion can be calculated based solely on the sensor signal 70. These distortion values ​​should then be minimized.

[0046] Second example: Alternatively, both the sensor signal 70 and the input signal 20 can be used to consider the relationship between the two signals, in particular the (in)coherence. Thus, the total non-coherent distortion (TNCD) can be calculated for the complex input and sensor signals, and then this must be minimized, for example.

[0047] The first objective function 80 is defined such that minimization results in a reduction of the nonlinear distortion portion at the output of the sink 50. When minimizing the objective function 80, the nonlinear predistortion at the output of the circuit 100, i.e., the sink 50, is reduced by predistorting the signal. A function that calculates a measurement that characterizes the nonlinearity of the system is suitable as the first objective function 80.

[0048] Preferably, controller 30 is configured to normalize predistorted signal 40 to keep predistorted signal 40 at the output of controller 30 within the original operating range of input signal 20, specifically, the original normalized operating range is −1≦x≦1. Usually the operating range is known, thereby allowing for its normalization.

[0049] The at least one sensor signal 70 of the sink 50 includes a measured output voltage, and may output current intensity and / or sound pressure and / or surface vibrations, and the like.

[0050] Preferably, the conditioning unit 60, as shown in FIGS. 1 and 2, is a DC voltage source that provides the conditioning signal as a DC voltage to the sink 50. The operating point and / or operating range are preset or adjusted by the conditioning signal provided at the sink 50. The term operating range, on the other hand, refers to the AC voltage operating range of the input signal and the predistorted signal, which may be, for example, between −1 V and 1 V. With respect to the AC voltage alone, this operating range at the sink is maintained but shifted by a DC offset resulting from the conditioning signal. This result may be, for example, an operating range between 9 V and 11 V for a 10 V conditioning signal. Thus, the conditioning signal influences the “positioning” or “stroke” of the operating range on the characteristic curve (e.g., ±1 V), which is initially preset by the input signal 20 and / or ultimately by the predistorted signal 40. Therefore, the operating range may be understood as a signal stroke or absolute value. Such a distinction is well known to those skilled in the art.

[0051] The adjustment signal, specifically an arbitrarily selected DC voltage, causes the sink 50 to have an operating point or operating range that is fixedly preset or that varies only slightly depending on the application. In this case, the term "slightly" should be understood to mean "gradually over time," meaning that the operating point changes so slowly that the predistortion parameter optimization, which is performed in parallel with normal operation, can (at least approximately) converge. The speed of convergence depends primarily on the optimization hyperparameters, the nonlinear characteristic curve of the sink, and the statistical characteristics of the input signal. When the operating point changes, the speed of convergence also depends on how strongly / by how much the operating point is shifted. Too rapid and continuous strong changes hinder convergence and therefore proper predistortion. Often, operating point changes occur both quickly and strongly at the same time (see above). This means that the operating point change cannot be continuously tracked, and therefore the predistortion parameter optimization or parameter variation 90 fails.

[0052] Preferably, the controller 30 is configured to modify the at least one predistortion parameter based on the at least one sensor signal 70 so that the first objective function 80 is minimized, in particular so that the first objective function 80 is calculated based on the at least one sensor signal 70, or so that the first objective function 80 is calculated based on the at least one sensor signal 70 and the input signal 20. As explained previously, the at least one predistortion parameter r may be a real number, or may be a vector or M-tuple (r...r M ) may be indicated.

[0053] Additionally, the first objective function 80 preferably includes one or more functions that determine one or more measurements that characterize the nonlinearity of the system. The measurements are, for example, distortion factor or total harmonic distortion (THD) or total noncoherent distortion (TNCD). Further measurements can be intermodulation distortion, THD+N (THD+Noise), generally cross-correlation based methods (TNCD, among others).

[0054] The first objective function 80 can be extended, for example, by a measure of level change, particularly a level decrease or increase, at the output of the sink 50. Therefore, when optimizing, a weighting can be made between the nonlinear distortion 40 and the level decrease. Predistortion can result in significant level changes, both upward and downward, depending in particular on the specific combination of the predistortion function and the sink's characteristic curve. To detect level changes in an input signal with a constant level over time, the output level with and without predistortion is detected to obtain a measure of the level change. The optimization can be based on data from a certain frequency range. This frequency range can be, for example, within the useful signal bandwidth, but can also be partially or entirely outside the useful signal bandwidth. An example of this is the ultrasonic range traditionally used in audio applications. The frequency range for audio or audible sound is generally considered to be from 20 Hz to 20 kHz. In nonlinear systems with characteristic curves that are at least largely frequency-dependent, signals for characterizing the system's nonlinearity can be obtained above 20 kHz, but in the ultrasonic range where they are no longer perceptible. This can also occur in addition to ordinary audio signals. The useful frequency range may be defined differently depending on the application, for example between 0.1 Hz and 10 Hz, in which case a frequency range above 10 Hz may be used, for example, to observe nonlinearities. For very low frequency audio applications, for example below 20 Hz, such test signals are also considered for frequency ranges below the useful frequency range. This method can be advantageous, since the useful frequency range may be less loaded or may not be loaded at all. Optimizing the first objective function 80 can be performed in different ways.Depending on the selected implementation of the first objective function 80 and the input signal 20, classical optimization methods for finding a local minimum in parameter space and along the time axis, such as gradient methods, may be used, or else more complex methods for finding a minimum in parameter space and along the time axis are required, which may be, for example, from the area of ​​embedded optimization.

[0055] The first objective function 80 can also be minimized by an extremum adjuster that is appropriately tuned for any input signal, where "appropriately" means that the hyperparameters of the extremum adjuster are tuned so that the probability of convergence for any application-relevant signal with any characteristics (quasi-static, impulse-like, speech or music-like, stochastic) is very high.

[0056] Optimization of the first objective function may occur, for example, once, continuously, at certain time intervals, or when a threshold value of the first objective function is exceeded.

[0057] Preferably, the controller 30 is configured to minimize the first objective function 80 using a mathematical optimization method. Specifically, the controller 30 additionally selects a mathematical optimization method based on the characteristics of the input signal, or is configured to minimize the first objective function using a suitably tuned extremum controller. Extremum controllers are a subgroup of optimization methods suitable for the present application, since they can be used online while minimizing the objective function with little noticeable adverse effect. Extremum controllers continuously estimate the gradient at each control variable operating point and correspondingly change the control variables to minimize or maximize the objective function 80. If the hyperparameters of the extremum controller are tuned appropriately, the tuning will converge for most different, varying input signals, which is very important for use during normal operation.

[0058] The input signal 20 may be taken into account in the objective function 80. As a result, the objective function 80 may exhibit a dependency on the input signal 20. The type of optimization method may be selected based on the characteristics of the input signal 20. For example, the selection of the optimization method may be automatic, at least to some extent. For example, if a static input signal 20, such as a continuous sinusoidal sound, is detected by the controller 30, a classical gradient method may be automatically selected. For a static input signal 20, the classical gradient method is sufficient. For example, if a changing / arbitrary input signal 20 is detected by the controller 30, a global optimization method, particularly a time-dependent method, such as embedded optimization, a genetic algorithm, or an extremum controller, may be automatically selected. Specifically, an extremum controller is global only with respect to time and local to the objective function.

[0059] Preferably, when the first objective function 80 includes a measure of the change in level at the output of the sink 50, the controller 30 is configured to weight the first objective function 80 with respect to the non-linear distortion and, in particular, optionally, the change in level at the output of the sink 50. The term change in level includes both a decrease in level and an increase in level, both of which may have undesirable effects, so the weighting of the objective function may be careful in this respect. The optional adaptation of the level, in particular the increase in level 39, may change the operating range, since the maximum and minimum values ​​40 and 40a of the predistorted signal 40 may subsequently exceed or fall below the range of input values, but the operating point is maintained.

[0060] Preferably, the controller 30 is configured to minimize the first objective function 80 based on at most N-1 repetitions of the predistortion iteration steps of different orders, where N is a natural number greater than 1. Preferably, the predistortion iteration steps are performed in ascending order. It is also possible to skip some orders or to start only with higher orders where n>2. For example, 2, 3, 4, 5 is an ascending order without skipping, and 2, 4, 5 is an ascending order with order 3.

[0061] The following examples explain in more detail what is intended by omitting certain orders or starting with higher orders. For example, sink 50 may primarily exhibit third- and sixth-order predistortion sections. While intuitively, the result is a predistortion cascade consisting of orders 3 and 6, it could also consist of orders 3, 4, 5, 6, 7, 8, ... or 3, 4, 6, 8, 9, 12, .... Different cascades are not strictly equivalent in terms of effectiveness, but they have similar effects when converging to the minimum of the first objective function. Depending on the characteristic curve to be compensated, it may be advantageous to consider many different orders in the cascade. It is important that the lower-order predistorted signal 40 is the input signal for the higher-order predistortion. Here, the cascade is always in ascending order. It is also noted here what is intended by different orders. For example, second harmonic distortion has twice the fundamental frequency (the first harmonic), third harmonic distortion has three times the fundamental frequency (the second harmonic), etc. Thus, in harmonic distortion, the order of distortion, n, indicates that the (n-1)th harmonic is involved.

[0062] 3 shows a schematic diagram of a controller 30 including possible components of a predistortion block 32 using an example of the first variant. The predistortion block 32 includes N-1 predistortion iteration steps, with parameter variations 90 being performed in each predistortion iteration step. Thus, the predistortion block 32 includes a predistortion cascade, which includes a second-order predistortion iteration step to an n-th-order predistortion iteration step, where N is a natural number greater than 1. FIG. 3 shows the cascade in ascending order without any omissions.

[0063] In other words, the predistortion block 32 of Figure 2 consists of a cascade of N-1 predistortions of different orders, starting with order 2 and ending with order N. Each predistortion should be understood to be an iteration step of the iteration. These predistortions or iteration steps of different orders are represented by functions v2 to v N The predistortions are arranged serially in increasing order, such that each predistortion order processes the signal predistorted with the next lower order. Expressed mathematically, the predistorted signal

number

number

number

[0064] Preferably, after performing the iterations, the controller 30 is configured to output the predistorted signal 40 and forward it to the sink 50, as shown in particular in the predistortion block 32, which in turn forwards the sensor signal 70 to the controller 30. Possible sinks with which the method works well are, among others, amplifier circuits, discrete electronic devices such as transistors, or electromechanical transducers such as dielectric elastomer actuators and electrostatic actuators. In the field of audio systems, all these types of sinks can be applied. However, the application of the method is not specifically limited to the audio field.

[0065] Figure 4 shows the nth-order predistortion n4 shows a schematic representation of possible components of the predistorted signal 40. The possible components include different functions that can be performed by the controller 30. For example, a sample rate enhancement 34 and a sample rate reduction 36 can be optionally performed. Specifically, the sample rate enhancement 34 is performed before an n-th order predistortion function 35, where n is a natural number greater than 1. In addition, the sample rate reduction 36 is performed, in particular, after the predistortion function 35 is applied. After the n-th order predistortion function 35, as shown in FIG. 4, a mean value adaptation, in particular a mean value suppression 37, is performed by or in the controller 30, followed by a normalization 38 of the predistorted signal 40, after which the predistorted signal 40 is forwarded to a sink 50. If the operating point is to be maintained, the DC portion of the predistorted signal 40 must be removed / suppressed, which is done by subtracting a calculated mean value and / or by a high-pass filter (wherein mean value suppression is performed, as shown in FIG. 4). However, if the operating range is to be maintained, the DC portion does not have to be completely removed but must be adapted in some way. Therefore, in this case, a value for adapting the DC portion of the predistorted signal is calculated (in this case, mean value adaptation is then performed, as shown in FIG. 10). The calculation of the mean value for removing / suppressing the DC portion and the calculation of the value for adapting the DC portion, and therefore the required mean value adaptation, can be calculated using the predistortion parameter r. In other words, both the mean value suppression and the mean value adaptation, and their respective (mean) values, can be calculated entirely based on the predistortion parameter r. Maintaining the operating point or operating range represents extreme values ​​between which movement can be performed by adapting the DC portion accordingly. Mean value adaptation 37 and normalization 38 together can also be understood as compensation for a DC offset. Mean value adaptation affects the DC portion of the predistorted signal 40. Normalization affects both the DC portion and the AC amplitude if the DC portion is still present after mean value adaptation.If a DC portion still exists, it is amplified by normalization, as is the AC amplitude. These two functions together result in maintaining or approximately maintaining the operating point or operating range. In the case of mean value suppression, which occurs when the operating point is to be maintained, mean value adaptation alone has the effect of "compensating for the DC offset." The DC offset is introduced into the predistorted signal 40 by the predistortion function 35. Optionally, a level change, specifically a level boost or a level reduction, can be performed before passing it on to the predistorted signal 40. For example, FIG. 4 optionally shows a level boost 39. The level boost 39 is an example of level adaptation 39. The mean value suppression 37 involves calculating an average value and subtracting the calculated average value from the predistorted signal 40. In addition, the mean value suppression can use a correspondingly adjusted high-pass filter, specifically a high-pass filter without bandwidth loss. However, mean value adaptation involves adapting the DC portion of the predistorted signal 40 to calculate a value that is then subtracted from the predistorted signal 40 .

[0066] As indicated by the arrows in Figure 4, the steps of predistortion 35, mean suppression 37, normalization 38, and optionally level enhancement 39 result in parameter variations 90. In steps 35, 37, and 38, one and the same set of parameters is used. Step 39, i.e., level compensation, uses an additional parameter g. However, parameter g may also be considered a predistortion parameter, r M These parameters may be considered to be parameters of parameters, which means that the entire predistortion block involving steps 34 to 39 uses one and the same set of parameters that are varied specifically for optimization.

[0067] As shown in Figure 4, different steps 34 to 39 are performed with each predistortion order 2 through N, as shown in predistortion block 32 of Figure 3. Figures 3 and 4 each show, in somewhat different detail, how controller 30 is configured to operate. Specifically, Figures 3 and 4 together show how controller 30 is configured to operate.

[0068] If the AC input signal 20 is not sufficiently band-limited to suppress most of the temporal aliasing, such aliasing, which may occur due to the following non-linear signal processing, may be avoided by an optional increase 34 in the sample rate, in particular according to step 35. The correspondingly band-limited signal is then processed by an n-th order pre-distortion function 35, which may include, but is not limited to: y=(a+bx) n is such that all the above nonlinearities of the form are at least approximately compensated for. Thus, the coefficients a and b, the output signal y and the input signal x are real numbers, and n≧2 is a natural number describing the order of the nonlinearity. A possible predistortion function that meets this objective can be described by the following equation:

[0069]

number

[0070] Therefore, the parameter r ≠ 0 and the signal being processed nonlinearly

number

number

number

number

[0071] As can already be inferred from this description, at least one distortion parameter includes several distortion parameters r. The distortion parameter r can in particular be understood as a vector quantity. The nth-order distortion function can have one or more parameters. As in the above example, the distortion parameter represented by r is, for example, a real number. In other functions, it can also be r1, r2, …. In addition, the strain parameter r as a vector quantity may include an optional g for level compensation, which is a further r k (k from 1 to K). In this case, each different order of distortion has K parameters. Then, the distortion block includes L (L is at least equal to 1 and at most N - 1) distortions each having K parameters. In this case, the total number of distortion parameters is M = K * L.

[0072] As also shown in FIG. 3, the controller 30 is configured to adapt each distortion parameter r of each of the L distortion iteration steps to the characteristics of the sink 50 either successively in ascending order of order by one-dimensional optimization or in parallel in time by multi-dimensional optimization of the first objective function 80, where 1 ≦ L < N. The number of iteration steps depends strongly on the maximum order N, since not all orders n (as already discussed) need to be included in the cascade.

[0073] Each of Figures 5 to 8 shows simulation results obtained using the proposed circuit 100 according to the method as shown in Figures 3 and 4. Figures 5 to 8 illustrate the basic effectiveness of predistortion. The nonlinearities on which Figures 5 to 8 are based to obtain the simulation results are depicted in the corresponding figures. The simulated amplitude in dB is plotted over the frequency in Hz, respectively. Each of Figures 5 to 7 shows, using simulation, the effect of the method on a simulated sensor signal 70 of a simulated nonlinear system, the characteristic curve of which is y=(1+0.75x). 2 and a 1 kHz sinusoidal tone with a peak value of 1 was used as the input signal 20. "No DSP" in the descriptions of Figures 5 to 8 means that the method was not applied to the simulated nonlinear system.

[0074] Thus, Figure 5 represents a simulated sensor signal 70 without the effect of this method (without DSP) compared to a simulated sensor signal 70 using this method (with DSP). Figure 5 shows a significant reduction in the amplitude of the second harmonic 86 by over 80 dB, while the amplitude of the fundamental frequency 88 is reduced by only about 2.5 dB.

[0075] 6 represents a simulated sensor signal 70 without the effect of the method (without DSP) compared to a simulated sensor signal 70 using the method (with DSP and with level compensation) that includes level compensation. FIG. 6 still shows a significant reduction in the second harmonic 86 of approximately 50 dB, but only a level reduction of about 1.5 dB at the fundamental frequency 88. It can be seen that additional higher order harmonics are generated, each roughly −60 dB in amplitude compared to the amplitude of the fundamental frequency 88. This illustrates the balance between the level reduction (at the fundamental frequency 88) and the nonlinear distortion in the sensor signal 70.

[0076] 7 shows the different amplitudes of the fundamental frequency 88 in a simulated sensor signal 70 without the effect of the method (without DSP), a simulated sensor signal 70 using the method without optional level compensation (with DSP), and a simulated sensor signal 70 using the method including level compensation (with DSP, including level compensation). The level reduction without DSP is by definition 0 dB, with DSP it is approximately 2.5 dB, and with DSP including level compensation it is approximately 1.5 dB.

[0077] Figure 8 shows the simulation result of the characteristic curve y=(0.5+0.375x) 2 +(0.25+0.5x) 3 1 shows the effect of the method on a simulated sensor signal 70 of a simulated nonlinear system where . Using the method (with DSP), a significant reduction of the second harmonic 86 by approximately 55 dB and the third harmonic 87 by approximately 25 dB can be seen. The level of the fundamental frequency 88 is reduced by approximately 5 dB. In addition, predistortion generates higher order harmonics, with the highest amplitude (fourth harmonic 89) being approximately -45 dB compared to the amplitude of the fundamental frequency 88.

[0078] Looking at Figures 5 to 8 together, it should be pointed out that the two simulated systems (Figures 5 to 7 and Figure 8) are very different (see quadratic for the characteristic curves of Figures 5 to 7 vs. cubic for the characteristic curve of Figure 8, and the different polynomial coefficients), with most of the different graph results already "without DSP".

[0079] In Fig. 5, a method with only second-order predistortion is applied, while in Fig. 8, a method with a cascade of second and third orders is applied. The resulting predistortion parameter r also differs due to these facts. The interaction between the respective nonlinear systems and the respective predistortions is consequently very different.

[0080] FIG. 9 shows the components of the control unit 30 of the second variant of the method. The second variant follows an adaptive control paradigm (adaptive feedforward control). In contrast to the first variant of the method, in the second variant, a method 100, which is substantially equivalent to the first variant, is virtually executed in the controller 30. The virtual components each have the same reference numeral as the real components, supplemented by the letter "a" (see FIGS. 9 and 10). Consequently, the virtual signal source has the reference numeral 10a, the real signal source has the reference numeral 10, and so on. A further difference is that the second objective function 85a is first virtually optimized by adapting the parameters of the nonlinear system model 50a so that the deviation between the real sensor signal 70 and the virtual sensor signal 70a is minimized. The second objective function 85a is virtually optimized because the optimization of the second objective function 85a is performed "on the fly" without interfering with the operation of the sink 50 with the input signal 20 or the predistorted signal 40. Once this nonlinear system has been identified, as in the first variant, the first objective function 80a is virtually optimized to minimize the nonlinear distortion at the output of the nonlinear system model 50a, while taking into account level degradation.

[0081] The advantage of this procedure is that the optimization process is almost entirely based on virtual signals and is therefore largely independent of the actual circuit 100. The optimization process has no impact on the standard operation of the circuit 100, which can continue in parallel. Therefore, the length of the optimization process is not critical. The optimization of the first objective function 80a can be accelerated compared to the first variant because it does not have to wait for the actual sensor signal 70. After completing all optimizations, i.e., once optimization 30b is finished, the predistortion 30a is switched between the actual AC signal source 10 and the actual sink 50 by transmitting the predistortion parameters to the control unit 30. Both the parameters of the nonlinear system model 50a and the parameters of the predistortion 30, 30a can be adapted, for example, once, continuously, at certain time intervals, or when a threshold value of the first objective function 80a or the second objective function 85a is exceeded.

[0082] Preferably, the controller 30, 30a is configured to minimize the second objective function 85a by adapting at least one model parameter of the nonlinear system model 50a so that the deviation between the actual sensor signal 70 at the output of the sink 50 and the virtual sensor signal 70a at the output of the nonlinear system model 50a, or between a quantity derived from the actual sensor signal 70 and a quantity derived from the virtual sensor signal 70a, is minimized. It should be pointed out that in general, the "predistortion parameters" parameterize only the predistortion, specifically only the n-th order predistortion. The "model parameters" are model parameters that adapt the nonlinear system model 50a to the actual sink 50. Thus, the model parameters also play a role in minimizing the second objective function 85a. The nonlinear system model may be realized and thus parameterized in many different ways. A simple purely nonlinear system model may be, for example, a system model with y = ax 2where the parameters are a, the input signal is x, and the output signal is y. Possible nonlinear system models 50a for the sink 50 therefore range from block-based models (such as the Hammerstein model) operating using pure polynomials or FIR / IIR filters via physically guided state-space models, to (deep) neural networks such as LSTM-NN. The more accurately the model can map reality, the better and more robust its performance will be when compensating for nonlinearities. However, generally speaking, the algorithms disclosed herein are independent of the model chosen.

[0083] The term "quantity derived therefrom" should be understood to mean minimizing not the (time-averaged) deviation between the measured sensor signal 70 (e.g., current intensity) and the virtual sensor signal 70a (e.g., current intensity based on a model), even if the directivity to the target is not necessarily equal, but also the deviation in the frequency range or the deviation of the respective measures of the nonlinearity derived from the measurement and the nonlinearity derived from the model prediction (THD, TNCD, ...).

[0084] Preferably, the controller 30, 30a is configured to perform the minimization of the second objective function 85a, followed by the minimization of the first objective function 80a as previously described for the first variant, to minimize the nonlinear distortion in the output of the nonlinear system model 50a. It should be pointed out that the optimization of the first objective function 80a and the optimization of the second objective function 85a are each performed virtually.

[0085] Preferably, the controller is configured to perform a virtual minimization of, among other things, the second objective function 85a and the first objective function 80a, and then output the actual predistorted signal 40 to be transmitted to the sink 50. By being able to virtually perform the minimization of the first objective function 80a and the second objective function 85a, this optimization process does not affect the correct operation of the circuit 100.

[0086] Figure 10 shows, by analogy with Figure 4, the nth-order virtual predistortion n 1 schematically illustrates possible components of the virtual controller 30a. The possible components include different functions that can be performed by the virtual controller 30a. For example, a virtual sample rate increase 34a and a virtual sample rate decrease 36a may be optionally performed. Specifically, the virtual sample rate increase 34a is performed before the virtual n-th order predistortion 35a, where n is a number greater than 1. In addition, specifically, the virtual sample rate decrease 36a is performed after the virtual predistortion 35a. After the virtual n-th order predistortion 35a, a virtual mean value adaptation 37a, specifically a virtual mean value suppression 37a, is performed by or in the virtual controller 30a, followed by a virtual normalization 38a of the virtual predistorted signal 40 and optionally a virtual level adaptation 39a, specifically a virtual level enhancement 39a, so that when the optimization ends 30b, the determined predistortion parameters 30c are transmitted to the controller 30, which then transmits the actual predistorted signal 40 together with the determined predistortion parameters 30c to the sink 50. Optionally, a level change 39, specifically a level enhancement or level reduction, can be performed before transmitting the actual predistorted signal 40. For example, FIG. 10 shows an optional level enhancement. The mean value suppression 37a includes calculating an average value and subtracting the calculated average value from the virtual predistorted signal 40a. Mean value adaptation involves determining a value to adapt the DC portion without eliminating it completely and subtracting this value from the mean value. As shown in Figure 10 by the arrows, parameter variations 90a are virtually performed in the steps of predistortion 35a, mean value suppression / mean value adaptation 37a, normalization 36a, and optional level enhancement 39a.

[0087] 10, two loops in which virtual parameter variations 90a and 90a' are performed can be seen. The first loop has a signal path of the virtual sink 50a (=nonlinear system model 50a) → virtual sensor signal 70a → virtual optimization of the second target function 85a → virtual parameter variation 90a' → virtual sink 50a.

[0088] The second loop in Figure 10 has a signal path from the virtual sink 50a (=nonlinear system model 50a) → the virtual sensor signal 70a → the first objective function optimization 80a → the virtual parameter variation 90a → the virtual controller 30a → the virtual predistorted signal 40a → the virtual sink 50a.

[0089] It should be noted that the description already given for the first variant can be directly transferred to the second variant. Specifically, embodiments that can be implemented using real components of the circuit 100 can also be implemented using virtual components of the circuit. The result is, for example, that at least one virtual predistortion parameter r can be a real number or a vector quantity (see the description of the first variant). The description of the first variant will not be repeated again for the second variant. Rather, for the second variant, reference is made here to the description of the first variant.

[0090] In the first and second variants, the actual input signal 20 may be split into frequency bands by a filter bank before predistortion, which may then be predistorted differently and then added back together to form the overall signal.

[0091] FIG. 11 and FIG. 12 each show simulation results for the second method presented herein, implemented using circuit 100. FIG. 11 shows, in FIGS. 11a and 11b, THD prediction and THD reduction versus frequency in Hz. FIG. 11 is subdivided into FIGS. 11a and 11b because, when plotted in a single drawing, they overlap and are indistinguishable from one another. FIG. 11a shows the THD values ​​of the original simulated nonlinear system (“Original”) without and with the second variant of the method (“DSP”) applied. The THD values ​​of the original simulated nonlinear system without DSP are between approximately 3% and 35%, and with DSP are between approximately 0.06% and 0.7%. This represents a reduction of 50 times, or approximately 34 dB.

[0092] FIG. 11b shows the THD values ​​of the hypothetical nonlinear system model ("Model") without and with the second variant of the method ("DSP"). The model without DSP almost identically predicts the THD values ​​of the original simulated nonlinear system (see FIG. 11a, graph "Original without DSP"). This result is obtained after optimizing the second objective function 85a. As might be expected, the reduction of nonlinear distortion is better with the model because the model deviates slightly from the original system, but the predistortion parameters were optimized using the model. The THD values ​​with DSP are generally between 0.04% and 0.4%. Essentially, FIG. 11 shows that applying the second variant of the method after all optimizations in FIGS. 11a and 11b results in a significant reduction of the nonlinear distortion of the simulated real sink 50, in this case.

[0093] FIG. 12 shows the transfer functions for the prototype and the model. The transfer function is expressed as a dependence of amplitude in dB on frequency in Hz. It can be seen that the model and the prototype are nearly superimposed one on top of the other. FIG. 12 shows that the hypothetical nonlinear system model ("model") after optimizing the second objective function 85a predicts with great accuracy (maximum deviation <0.5 dB) the transfer function of the original simulated nonlinear system ("prototype"), which is unknown from the model's perspective. This, in combination with FIG. 11, shows that the nonlinear system model 50a after optimizing the second objective function 85a can map the linear and nonlinear transfer characteristics of the actual sink 50, in this case the simulated one.

[0094] Figures 11 and 12 show that the second variant works in terms of both system identification (adapting a virtual system model to the sink) and predistortion (applying virtually optimized parameters to predistortion for the real sink).

[0095] It should be pointed out here that the present disclosure describes predistorted signals 40, 40a. By feeding back the sensor signal 70 of the sink 50 to the controller 30 and continuously adapting the predistortion parameter r in a loop, the predistortion can be continuously adapted so that, in the best case, the second signal 70 no longer contains any distortion after the expiration of the time interval in which the optimization was terminated. At the same time, the operating point and / or operating range of the circuit can be operated by adjusting the signal at an essentially constant operating point and / or constant operating range.

[0096] FIG. 13 shows simulation results of adaptive predistortion when the operating point is not changed. In FIG. 13, the DC operating point 71, the calculated values ​​of the predistortion parameters 73, and the total noncoherent distortion (TNCD) 72 are shown. It can be seen that the predistortion parameters 73, starting from their initial values, adapt to the DC operating point 71 between 0 s and 140 s, while the total noncoherent distortion 72 decreases to a value of 0. Between 140 s and 250 s, the operating point 71, the calculated predistortion parameters 73, and the total noncoherent distortion 72 remain constant. Additionally, FIG. 13 shows that when the DC operating point 71 is moved, the total noncoherent distortion 72 increases sharply at approximately 250 s. As a result, the predistortion parameters 32 are set to a shifted DC operating point 71, which results in a reduction in total non-coherent distortion 72 (TNCD), particularly from 250 s to 300 s. From approximately 300 s onwards, the DC operating point 71, calculated predistortion parameters 73, and total non-coherent distortion 72 remain constant.

[0097] Another aspect of the present disclosure relates to a method for compensating for nonlinearity without essentially changing the operating point and / or operating range of the characteristic curve. Figure 14 shows a flowchart of a proposed method 130. The method 130 comprises at least steps 131 to 136. In step 131, the method 130 comprises providing an input signal by an AC voltage signal source.

[0098] Specifically, the AC voltage signal provided by the AC voltage signal source in this case should be understood to be discrete amplitude values ​​at regular sampling times, which reach the controller. However, it is also conceivable that the controller is provided with an analog signal that can be converted into a digital signal by the controller. An analog implementation of the controller 30 is fundamentally possible, and analog signals can also be used. However, digital or digitized signals are preferably used. Here, A / D conversion can be performed before the controller 30 and D / A conversion can be performed after the controller 30.

[0099] In step 132, the method 130 includes receiving an input signal 20 by the controller 30 and converting the input signal 20 into a predistorted signal 40 using at least one preset predistortion parameter. The at least one preset predistortion parameter may be stored in a database accessible by the controller 30. Thus, if requested by a user or after completing a previous optimization, the at least one preset predistortion parameter may be saved, specifically, overwritten and / or stored in the database. A discussion of the predistorted signal 40 has already been given when describing the circuit. To avoid redundancy, this discussion will not be repeated.

[0100] In step 133 , the method 130 comprises subsequently receiving the predistorted signal 40 by a sink 50 , which is coupled to a conditioning unit 60 .

[0101] In step 134, method 130 includes, concurrently with receiving the predistorted signal by the sink, providing, by the adjustment unit, an adjustment signal to the sink to operate the sink within an operating range or at an operating point. By providing the adjustment signal to sink 50, the operating point and / or operating range of circuit 100 may be adjusted.

[0102] The operating point (DC offset) is predetermined by adjustment signal 20. If the operating point is intentionally or unintentionally changed, the methods disclosed herein are able to track the changes fast enough (with respect to optimizing predistortion parameters) when the changes are only slight and / or occur relatively slowly. This was discussed in detail above with respect to circuit 100, and reference is made again here.

[0103] In step 135, the method 130 continues with receiving at least one sensor signal output at the sink in a feedback manner by the controller for adapting at least one preset predistortion parameter based on the at least one sensor signal.

[0104] At least one predistortion parameter is frequently adapted until the distortion of the sensor signal 70 is compensated for in the best possible way, where the input signal is predistorted with each parameter variation for each current parameter, specifically according to the control loop described herein. Compensation for the distortion is automatic.

[0105] In step 136, the method 130 includes converting the input signal into a predistorted signal by at least one adapted predistortion parameter to provide the predistorted signal to the sink without essentially changing the operating point of the characteristic curve and / or the operating range of the characteristic curve. To maintain the operating point or range of the characteristic curve, the input signal 20 itself is not changed.

[0106] Method steps 131 to 136 are preferably performed one after the other in ascending numerical order, with steps 133 and 134 being particularly preferably performed in parallel. Steps 131 to 136 are all parallel. At first glance, this may seem contradictory, but it is not. This is because in an analog implementation, they occur continuously simultaneously. However, in a preferred digital implementation, they are performed simultaneously at fixed time transitions, such as the sample rate. In particular, in a digital implementation, some steps may take somewhat longer than others. For example, an input signal may be received at every sampling time and a predistorted signal may be output at every sampling time. However, optimization of the first objective function may be performed based on a signal block consisting of P sample values, with parameter variations occurring only at each of the P sampling times, where P is a natural number. For example, a signal block may contain P=256 sample values. A signal block may also contain a different number of sample values. Nevertheless, the steps are performed in parallel.

[0107] Preferably, method 130 comprises the step of providing a regulation signal in the form of a DC voltage to sink 50 by regulation unit 60, with the regulation signal being an arbitrary selected DC voltage, in particular so that sink 50 has an operating point and / or operating range that is fixedly predetermined or that varies only slightly in relation to the application. The slight change in operating point and / or operating range comprises a time window having sufficient time for the optimization of the objective function to converge. This is discussed in more detail above with respect to circuit 100, to which reference is made again here.

[0108] Preferably, the method 130 comprises a step of varying the at least one pre-distortion parameter r based on the at least one sensor signal 40 so as to minimize the first objective function 80, in particular so that the first objective function 80 is calculated based on the at least one sensor signal 40 or so that the first objective function 80 is calculated based on the at least one sensor signal 40 or so that the first objective function 80 is calculated based on the at least one sensor signal 40 and the input signal 20. As a result, the first objective function 80 may exhibit a dependency on the sensor signal 40 or on the sensor signal 40 and the input signal 20. It is also conceivable to let the user decide whether the adaptation of the at least one distortion parameter should be performed using both signals 20, 40.

[0109] Preferably, the first objective function 80 includes one or more functions that determine one or more measurements that characterize the nonlinearity of the system, for example, the distortion factor, total harmonic distortion, or total incoherent distortion. With regard to the robustness of the optimization, it may be advantageous for multiple measurements to be considered in the first objective function 80. For example, if several sensor signals are available, these may be considered and weighted using the measurements. It may be advantageous to weight using several different measured distortion values.

[0110] Preferably, the method 130 comprises a step of weighting the first objective function 80 with respect to nonlinear distortion and level changes at the output of the sink 80, if the first objective function 80 includes a measure of the level change at the output of the sink 80. Essentially, the output level without predistortion will be compared with the output level while using predistortion, and the latter will be adapted to the former. In other words, the difference between the output level without predistortion and the output level with predistortion should be minimized, or else the ratio should be close to 1. According to a first option, the difference may be calculated, and according to a second option, the ratio may be formed. Additionally, it is conceivable to use the level (in dB, i.e., logarithmic) or to output the amplitude (in volts, for example, i.e., linear).

[0111] The respective output levels (with / without predistortion) can be detected by turning predistortion on / off or by determining the affected / unaffected parameters. The output levels should be equivalent. This is the case, for example, when the input signal has a constant characteristic over time, as is the case for quasi-static individual or multi-tone signals. In a second variant, more complex / dynamic signals can be used, since the virtual input signal is fully controllable and therefore reproducible.

[0112] Weighting the first objective function f1 may be expressed as: f1=A*(distortion measure)+(1-A)*(level change) where A is a real number, 0≦A≦1. The change in the scale and level of distortion can be weighted in f1 using the parameter A.

[0113] Examples of the first and second objective functions are given by the following functions:

[0114] 1. Objective function f1 f1 = A * THD(y) + (1 - A) * ΔL A is as described above, THD(y) is the total harmonic distortion in the sensor signal y at a certain frequency, and ΔL is the difference in the level of magnitude between y with predistortion and y without predistortion.

[0115] 2. Objective function f2 f2=(y - y virtual ) 2 That is, it is the squared deviation between the measured sensor signal 70 and the virtual signal 70a, which is typically averaged over a certain time frame.

[0116] Preferably, method 130 includes the step of minimizing the first objective function 80 by a mathematical optimization method, which is particularly selected based on the nature of the input signal 20, or the step of minimizing the first objective function 80 by an appropriately adjusted extremum regulator for any input signal 20. The selection of the optimization method has already been described in the context of the circuit and should be referred to here. The optimization of the first objective function should be performed either continuously in time or in parallel for different orders of predistortion.

[0117] Method 130 includes the step of performing N - 1 predistortion iteration steps of different orders n where 1 < n ≦ N to minimize the first objective function 80, and N is a natural number greater than 1. The optimization of the predistortion parameters can be performed in parallel for all predistortions. The predistortions must be applied to the signal one after another in ascending order of the order. The selection of the order is preferably done before the iteration. Some orders of distortion are identified as particularly important, either manually or automatically (for example, the third and fifth harmonics have particularly high energy, for example, they have a level of the fundamental frequency greater than 1% in a sine wave tone input signal), and then these orders (such as 3 and 5) are introduced into the distortion cascade. All other orders are not included in the cascade.

[0118] The order of predistortion used is, in particular, 1. To save time / reduce computational complexity 2. Additionally, potentially unwanted predistortion of different orders can unintentionally introduce further distortion at the output of the sink. For two reasons, we are limited to the smallest possible number of relevant degrees.

[0119] Therefore, preferably, the method 130 comprises a step of adapting the predistortion parameters of each of up to N-1 predistortion iteration steps to the characteristics of the sink, either successively in time in ascending order of n or in time parallel by multidimensional optimization of a first objective function.

[0120] Preferably, the method 130 comprises, after performing the iterations, outputting the predistorted signal 40 by the controller 30 for transmitting the predistorted signal 40 to the sink 50, and in particular for transmitting the sensor signal 70 to the controller 30. As a result, the sensor signal 70 is fed back to the controller 30, where the distortion parameter or at least one distortion parameter can be adapted if required.

[0121] Preferably, the method 130 includes a step of avoiding temporal aliasing by increasing the sample rate 34 of the input signal 20 if the input signal 20 is not sufficiently band-limited. As shown in FIGS. 4 and 10 , the sample rate increase 34 is performed in the controller 30, particularly after transmitting the input signal 40 to the controller 30. It can be inferred from FIG. 4 that the input signal 20, before increasing the sample rate using the sensor signal 70, enters the first target function 80. To calculate the first target function 80, the input signal 20 and the sensor signal 70 require identical sampling, and typically do. As a result, the sensor signal 70 never needs to be oversampled (see FIG. 4 ). Instead of or in addition to increasing the sample rate, the input signal 20 may be band-limited at this point.

[0122] Preferably, the method 130 calculates y=(a+bx) n where a and b are real coefficients, the output signal y and the input signal x are real, and n≧2 is a natural number describing the order of the nonlinearity. The range of values ​​−1≦x≦1 indicates the operating range of the input signal 40 being normalized. It is important that the operating range of the input signal 40, i.e., the maximum and minimum values ​​allowed, is known in order to maintain the operating range in future normalizations of the predistorted signal 40.

[0123] In a preferred embodiment, the predistortion function is such that it specifically compensates for a single order of distortion rather than simultaneously compensating for several orders of distortion. In other words, an nth-order predistortion includes an nth-order predistortion function and primarily compensates for nonlinear nth-order distortion. As a result, the equation y=(a+bx) n describes a type of nonlinearity that can be particularly well compensated by a single nth-order predistortion. Essentially, any, i.e., any complex predistortion function may be selected to at least partially compensate for nonlinearities of different orders, which nonlinearities may be, for example, represented by the polynomial y=a0+a1*x+a2*x. 2+... The nth-order predistortion includes an nth-order predistortion function and primarily compensates for nonlinear nth-order predistortion. An example of nonlinearity is y=(a+bx) n which has already been explained before.

[0124] A further preferred embodiment according to Fig. 3 describes, for example, the cascading or repetition of predistortions of different orders, which are imposed one after the other, making it possible not only to compensate for distortions of a single order, but in particular to compensate for distortions of different orders simultaneously. Thus, y = a0 + a1 * x + a2 * x 2 +…or y=(a1+b1*x) 2 +(a2+b2*x) 3 It is also possible to compensate for nonlinear distortions in more complex systems such as

[0125] Since the predistorted signal 40 in the first variant is continuously passed to the sink 50, compensation for nonlinearities is always performed, regardless of whether the first objective function 80 is being minimized or not. The result is that nonlinear distortions are minimized in general, and no particular form of nonlinearity is compensated for.

[0126] If sample rate enhancement 34 is performed before the nth predistortion iteration step 35, then sample rate reduction 36 is performed after the nth predistortion iteration step 35, specifically to return the input signal 20 to its original sample rate. Sample rate enhancement 34 and sample rate reduction 36 are both performed one after the other when the sample rate is changed. Sample rate enhancement 34 and sample rate reduction 36 are optional.

[0127] Preferably, the method 130 includes a step of detecting a DC offset in the signal, particularly in the predistorted signal 40, 40a, caused by the n-th order predistortion iteration step 35 (predistortion function 35, 35a). After detecting the DC offset, the DC offset is subsequently modified or removed. The DC offset removal can be performed, in particular, by a high-pass filter with a sufficiently low cutoff frequency and / or an average value calculation using the predistortion parameter r, followed by subtraction. The DC offset modification can be performed by adapting the DC offset. When removing the DC offset, the operating point is maintained. When adapting, the operating range is maintained. Here, the term "sufficiently" means maintaining the bandwidth of the input signal 20. Those skilled in the art will understand that, in terms of order, first the average value calculation is performed, then the subtraction, followed by the use of a high-pass filter. In FIG. 4, for example, the average value calculation is illustrated using the predistortion parameter r and subsequent subtraction, e.g., by the average value suppression 37. In FIG. 4, the average value suppression is used as an example of the average value adaptation 37. The DC offset, i.e., the direct current voltage portion, is an unwanted by-product of nonlinear distortion if the operating point is to be maintained. The DC offset is undesirable because it changes the operating point. Therefore, it is specifically proposed here to remove the DC offset again, and in particular to compensate for the DC offset, as will be explained.

[0128] Preferably, method 130 includes a step 38 of normalizing predistorted signal 40 to keep predistorted signal 40 at the output of controller 30 within the original operating range of the input signal. By compensating for the DC offset, the operating point and / or operating range can be kept essentially constant. If the operating range of input signal 20 is known, the operating range can be maintained by normalization.

[0129] According to a second variant of the method 130, the second objective function 85a is first minimized by adapting the predistortion parameters of the nonlinear system model 50a so as to minimize the deviation between the sensor signal 70 and the virtual sensor signal 70a at the output of the sink 50, or between a quantity derived from the sensor signal 70 and a quantity derived from the virtual sensor signal 70a. The minimization of the second objective function 85a is performed virtually, as already described for the first variant, so that the operation of the circuit 100 can continue independently. The second variant of the method is illustrated in FIGS. 9 and 10. Virtually, in this case, means that the actual sensor signal 70 is intercepted at least once, but more frequently in particular, and the virtual sensor signal 70a, corresponding to the output signal of the nonlinear system model 50a, is matched as closely as possible to the actual sensor signal 70 by optimizing the model parameters. The minimization of the second objective function 85a is performed “on the fly.” In particular, this does not interfere with the operation of the sink 50 with the input signal 20 or the predistorted signal 40 .

[0130] After minimizing the second objective function 85a, the first objective functions 80, 80a are minimized, as previously described in the context of the first variant of the method, to minimize nonlinear distortion in the output of the nonlinear system model. Consequently, the second variant of the method is initially performed virtually. The virtual execution may include minimizing the second objective function 85a and minimizing the first objective functions 80, 80a. The first objective functions 80, 80a are optimized, specifically, virtually minimized (second variant) or actually minimized (first variant). In the "adaptive feedforward" paradigm according to the second variant, the first objective function 80a is always optimized virtually. For example, in FIGS. 9 and 10, it can be seen that the actual sensor signal 50 is used only in optimizing the second objective function 85a, but not in optimizing the first objective function 80a. According to the second variant, preferably all objective functions are virtually optimized. Hybrid systems (a mix of the first and second variants) are also conceivable, which use a virtual system model on the one hand, but may also use the real sensor system 50 to optimize the first objective function 80a.

[0131] Additionally, after minimizing the second objective function 85a and the first objective function 80, 80a, the predistorted signal 40, 40a is output and transmitted to the sink 50, 50a. This can be done virtually or in reality. However, it is also conceivable that this procedure is first performed virtually and then in reality.

[0132] The second variant can be summarized as forming three steps as follows:

[0133] 1. Initial state: The real and virtual predistortions are adjusted to have no noticeable effect on the sensor signals 70, 70a, and the nonlinear system model 50a is still in the initial state. 2. A virtual optimization of the second objective function 85a and subsequently of the first objective function 80a. 3. After optimization, the real predistortion has the same parameters as the virtual predistortion.

[0134] Preferably, the adaptation of at least one parameter of the system model and / or at least one distortion parameter is performed once, continuously, at time intervals, or when a threshold of the first or second objective function is exceeded. Adapting relates to both a first variant, which particularly describes a closed-loop (feedback) variant of the method 130, and a second variant, which particularly describes an open-loop (feedforward) variant of the method 130. In the second variant, only "virtual" processing may be performed, since there is no closed-loop for the predistortion and the actual sink 50.

[0135] Preferably, in the method according to the first or second variant, the input signal 20, 20a, which is applied to the circuit 100, in particular, real and / or virtual, is divided into multiple frequency bands by a filter bank, and then the input signal (20, 20a) is transformed into a number of frequency-band-dependent input signals (20, 20a), followed by a predistortion 35, 35a, in particular real and / or virtual, of the frequency-band-dependent input signal 20, 20a. The number of frequency-band-dependent distorted signals 40, 40a are then combined to form a total predistorted signal 40, 40a, which is then transmitted to a sink 50 or a nonlinear system model 50a at the output of the controller 30, 30a. Frequency bands may need to be predistorted differently, resulting in different predistortions. A predistortion specifically optimized for each frequency band is then performed. Not all frequency bands should be distorted equally, except that the nonlinearities in all frequency bands behave identically.

[0136] A further aspect of the present invention relates to a computer-readable storage medium having stored thereon instructions that, when executed by a computer coupled to circuit 100, cause the computer to perform a first or second method as described herein using circuit 100.

[0137] It should be noted that analog implementations of circuits do not involve digital signal processors. Rather, analog passive and active electronic devices are used, specifically resistors (R), inductances (L), capacitances (C), operational amplifiers, diodes, transistors, potentiometers, etc. The signals being processed are analog, not digital. Analog implementations of circuits are contemplated.

[0138] While some aspects are described in the context of a device, it is understood that these aspects also represent descriptions of corresponding methods, and therefore device / circuit blocks or structural components should also be understood as corresponding method steps or features of method steps. Presenting the invention in terms of method steps is refrained from for the sake of redundancy. Some or all of the method steps may be performed by (or using) a hardware apparatus such as a microprocessor, a programmable computer, or an electronic circuit. In some embodiments, some or some of the most significant method steps may be performed by such an apparatus.

[0139] In the above detailed description, different features are grouped together in parts in the examples to streamline the disclosure. This type of disclosure should not be interpreted as intending the claimed examples to include more features than are expressly recited in each claim. Rather, as reflected by the following claims, subject matter may include fewer than all features of each disclosed example. Consequently, the following claims are incorporated into the Detailed Description, with each claim standing on its own as a separate example. While a dependent claim among the claims may refer to a specific combination with one or more other claims, it should be noted that other examples also include combinations of the dependent claim with the subject matter of each other dependent claim, or combinations of each feature with other dependent or independent claims. Such combinations are not intended unless it is expressly expressed that a specific combination is intended. In addition, a combination of features from one claim with each other independent claim may be used even if that claim is not directly dependent on that independent claim.

[0140] Depending on specific implementation requirements, intended embodiments of the present invention can be implemented in hardware, or in software, or at least partly in hardware, or at least partly in software. Implementations can be effective as long as they use digital storage media, such as floppy disks, DVDs, Blu-ray disks, CDs, ROMs, PROMs, EPROMs, EEPROMs, or flash memories, hard disks, or any other magnetic or optical memory, on which are stored electrically readable control signals that can cooperate or cooperate with a programmable computer system to execute the respective methods. This is why the digital storage media capable of carrying out the proposed teachings can be computer-readable.

[0141] Thus, embodiments in accordance with the invention described herein include a data carrier containing electronically readable control signals capable of cooperating with a programmable computer system to cause any of the methods described herein to be performed.

[0142] In general, embodiments of the teachings described herein may be implemented as a computer program product having program code that is effective to perform any of the methods when the computer program product is run on a computer.

[0143] The program code may also for example be stored on a machine readable carrier.

[0144] Other embodiments include a computer program for performing any of the features described herein as a method, the computer program being stored on a machine readable carrier. In other words, a method embodiment of the inventive method is therefore a computer program having a program code for performing any of the methods described herein, when the computer program runs on a computer.

[0145] A further embodiment of the proposed method is therefore a data carrier (or digital storage medium or computer readable medium) having recorded thereon a computer program for performing any of the methods described herein. The data carrier or digital storage medium or computer readable medium is typically tangible and non-volatile.

[0146] A further embodiment of the inventive method is, therefore, a data stream or a sequence of signals representing the computer program for performing any of the methods described herein, The data stream or sequence of signals may for example be adapted to be transmitted via a data communication link, for example the Internet.

[0147] A further embodiment comprises a processing means, for example a computer, or a programmable logic device configured to or adapted to perform any of the methods in the systems described herein.

[0148] A further embodiment comprises a computer having installed thereon the computer program for performing any of the methods described herein.

[0149] Further embodiments according to the invention include a device or system configured to transmit a computer program for performing at least one of the methods described herein in the form of a method to a receiver. This transmission may be, for example, electronic or optical. The receiver may be, for example, a computer, a mobile device, a memory device, or a similar device. The device or system may, for example, include a file server for transmitting the computer program to the receiver.

[0150] In some embodiments, a programmable logic device (e.g., a field programmable gate array, FPGA) may be used to perform some or all of the functions of the methods and devices described herein. In some embodiments, the field programmable gate array may cooperate with a microprocessor to perform the methods described herein. In general, the methods are performed in some embodiments by any hardware device, which may be any generally applicable hardware such as a computer processor (CPU) or method-specific hardware such as an ASIC.

[0151] The above-described embodiments merely represent exemplification of the principles of the present invention. It is understood that those skilled in the art will recognize modifications and variations of the arrangements and details described herein. This is why it is intended that the present invention be limited only by the scope of the following claims and not by the specific details presented herein by way of the description and discussion of the embodiments. [Explanation of symbols]

[0152] 10 AC signal source 20 Input Signals 30 Control Unit 32 Predistortion Blocks 34 Increased sample rate 35 nth-order predistortion functions 36 Sample Rate Reduction 37 Mean value suppression 38 Normalization 39 Level Up 40 Predistorted Signal 50 Sink 60 Adjustment Unit 70 Sensor Signal 71 DC operating point 72 Total noncoherent distortion 73 Predistortion Parameters 80 First Objective Function 90 Parameter Variations 100 circuits

Claims

1. A circuit (100) for compensating for nonlinearities without essentially changing the operating point of a characteristic curve and / or the operating range of a characteristic curve, comprising: an AC voltage signal source (10) for providing an input signal (20); a control unit (30) that receives the input signal (20) and converts the input signal (20) into a predistorted signal (40) according to at least one preset predistortion parameter; a sink (50) for receiving the predistorted signal (40), the sink (50) being coupled to an adjustment unit (60) configured to provide an adjustment signal to the sink (50) to operate the sink (50) within an operating range or at an operating point; Equipped with the control unit (30) is configured to receive at least one sensor signal (70) of the sink (50) in a feedback manner and to adapt the at least one preset predistortion parameter based on the at least one sensor signal (70); the control unit (30) converts the input signal (20) into a predistorted signal (40) by the at least one adapted predistortion parameter in order to provide the predistorted signal (40) to the sink (50) without essentially changing the operating point of the characteristic curve and / or the operating range of the characteristic curve; the control unit (30) is configured to modify the at least one predistortion parameter based on the at least one sensor signal (70) so that a first target function (80) is minimized, in particular so that the first target function (80) is calculated based on the at least one sensor signal (70) or based on the at least one sensor signal (70) and the input signal (20); Circuit (100).

2. 2. The circuit of claim 1, wherein, if the predistorted signal includes a DC portion, the control unit is configured to adapt or remove the DC portion, in particular, when removing the DC portion, by calculating the DC portion using the predistortion parameters and subsequently subtracting the calculated DC portion from the predistorted signal and / or by removing the DC portion from the predistorted signal by a high-pass filter with a sufficiently low cutoff frequency.

3. 3. The circuit (100) of claim 1 or 2, wherein the control unit (30) is configured to normalize the predistorted signal (40) to maintain the predistorted signal (40) at the output of the control unit (30) within an original operating range of the input signal (20).

4. 4. The circuit (100) according to any one of claims 1 to 3, wherein the at least one sensor signal (70) of the sink (50) comprises a measured output voltage and / or output current intensity and / or sound pressure and / or surface vibrations.

5. 5. The circuit (100) of any one of claims 1 to 4, wherein the conditioning unit (60) is a DC voltage source and provides the input signal as a DC voltage for the sink (50).

6. 6. A circuit (100) according to any one of claims 1 to 5, wherein the regulation signal, in particular any selected DC voltage, causes the sink (50) to have a predetermined operating point or operating range or to have an operating point or operating range that varies only slightly depending on the application.

7. 10. The circuit of claim 1, wherein the first objective function comprises one or more functions that determine one or more measurements for characterizing a nonlinearity of a system.

8. 8. The circuit (100) of claim 7, wherein the measurement is a distortion factor or a total harmonic distortion or a total non-coherent distortion.

9. The circuit (100) according to claim 1 or 7 or 8, wherein the control unit (30) is configured to weight the first objective function (80) with respect to the non-linear distortion and the change in the level at the output of the sink (50) when the first objective function (80) includes a measure of the change in the level at the output of the sink (50).

10. The control unit (30) is configured to minimize the first objective function (80) based on iterations of up to N - 1 predistortion iteration steps of different orders 1 < n ≦ N, where N is a natural number greater than 1. The circuit (100) according to any one of claims 1 or 7 to 9.

11. The control unit (30) is configured to minimize the first objective function (80) by means of a mathematical optimization method, specifically additionally configured to select the mathematical optimization method based on the nature of the input signal (20), or configured to minimize the first objective function (80) by means of an appropriately adjusted extremum regulator for any input signal (20). The circuit (100) according to any one of claims 1 or 7 to 10.

12. The control unit (30) is configured to output a predistorted signal (40) after performing the iterations and transmit the predistorted signal (40) to the sink (50), and the sink (50) is particularly configured to transmit the sensor signal (70) to the control unit (30). The circuit (100) according to claim 10.

13. The control unit (30) is configured to adapt the predistortion parameters of each of the predistortion iteration steps, particularly each of up to N - 1 predistortion iteration steps, to the characteristics of the sink (50) either in ascending order and temporally continuously by one-dimensional optimization of the first objective function (80) or temporally in parallel by multi-dimensional optimization of the first objective function (80). The circuit (100) according to claim 10 or 12.

14. 14. The circuit (100) of claim 1, wherein the control unit (30) is configured to minimize a second objective function (85a) and minimize nonlinear distortion at the output of the nonlinear system model by adapting at least one model parameter of a nonlinear system model such that a deviation between the sensor signal (70) or a quantity derived from the sensor signal (70) at the output of the sink (50) and a virtual sensor signal (70a) or a quantity derived from the virtual sensor signal (70a) is minimized before minimizing the first objective function (80, 80a).

15. 15. The circuit (100) of claim 14, wherein the control unit (30) is configured to output a predistorted signal (40, 40a) after performing the minimization of the second objective function (85a) and the first objective function (80, 80a) and to transmit the predistorted signal (40, 40a) to the sink (50).

16. 1. A method (130) for compensating for nonlinearities without essentially changing the operating point of a characteristic curve and / or the operating range of a characteristic curve, comprising: providing an input signal (20) by an AC voltage signal source (10); receiving said input signal (20) by a control unit (30); converting said input signal (20) into a predistorted signal (40) according to at least one preset predistortion parameter; Subsequently, receiving the predistorted signal (40) by a sink (50), the sink (50) being coupled to an adjustment unit (60); receiving the predistorted signal (40) by the sink (50) and simultaneously providing an adjustment signal by the adjustment unit (60) to the sink (50) to operate the sink (50) within an operating range and / or at an operating point; subsequently receiving at least one sensor signal (70) output at the sink (50) in a feedback manner by the control unit (30) for adapting the at least one preset predistortion parameter based on the at least one sensor signal (70); converting the input signal (20) into a predistorted signal (40) by the at least one adapted predistortion parameter to provide the predistorted signal (40) to the sink (50) without essentially changing the operating point of the characteristic curve and / or the operating range of the characteristic curve; modifying the at least one predistortion parameter based on the at least one sensor signal (70) so that a first objective function (80) is minimized, specifically so that the first objective function (80) is calculated based on the at least one sensor signal (70) or based on the at least one sensor signal (70) and the input signal (20); A method (130) comprising:

17. 17. The method (130) of claim 16, comprising providing the regulated signal (20) in the form of a DC voltage for the sink (50) by the regulating unit (60), in particular so that the sink (50) has a fixed, predetermined operating point or operating range, in particular by the regulated signal being any selected DC voltage, or has an operating point or operating range that varies only slightly depending on the application.

18. 17. The method of claim 16, wherein the first objective function comprises one or more functions that determine one or more measurements for characterizing nonlinearities of the system.

19. 19. The method (130) of claim 18, wherein the measurement is distortion factor or total harmonic distortion or total non-coherent distortion.

20. 19. The method (130) of claim 16 or 18, further comprising the step of weighting the first objective function (80) with respect to non-linear distortion and the change in level at the output of the sink (50), if the first objective function (80) includes a measure of the change in level at the output of the sink (50).

21. Minimizing the first objective function (80) by a mathematical optimization method, particularly selected based on the nature of the input signal (20), or minimizing the first objective function (80) by an appropriately adjusted extremum regulator for any input signal (20), the method (130) according to any one of claims 16 or 18 to 20.

22. Executing up to N - 1 predistortion iteration steps of different orders 1 < n ≦ N to minimize the first objective function (80), where N is a natural number greater than 1, the method (130) according to any one of claims 16 or 18 to 21.

23. After performing the iterations, outputting the predistorted signal (40) by the control unit (30), transmitting the predistorted signal to the sink (50), and particularly transmitting the sensor signal (70) to the control unit (30), the method (130) according to claim 22.

24. Adapting each of the predistortion parameters of the up to N - 1 predistortion iteration steps to the characteristics of the sink (50) either sequentially in ascending order in time or in parallel in time by multi - dimensional optimization of the first objective function (80), the method (130) according to claim 22 or 23.

25. When the input signal (20) is not sufficiently band - limited, avoiding temporal aliasing by improving the sampling rate (34) of the input signal (20), the method (130) according to any one of claims 22 to 24.

26. y=(a+bx) n 26. The method (130) of any one of claims 22 to 25, comprising the step of compensating for nonlinearities of the form: where a and b are real coefficients, the output signal y and the input signal x are real, and n≧2 is a natural number describing the order of the nonlinearity.

27. If the improvement of the sampling rate (34) is performed before the n - th predistortion iteration step, after the n - th predistortion iteration step (35), particularly performing a reduction of the sampling rate (36) to return the input signal (20) to the original sampling rate, the method (130) according to any one of claims 22 to 26.

28. Detecting the DC component in the predistorted signal (40) caused by the n - th predistortion iteration step (predistortion function), and subsequently modifying or removing said DC portion by a high-pass filter with a sufficiently low cut-off frequency and / or by averaging (37) using a predistortion parameter r and subsequent subtraction; 28. The method (130) of any one of claims 22 to 27, comprising:

29. 18. The method (130) of claim 16 or 17, comprising the step (38) of normalizing the predistorted signal (40) to maintain the predistorted signal (40) at the output of the control unit (30) within the original operating range of the input signal.

30. 18. The method (130) of claim 16 or 17, comprising the step of first minimizing a second objective function (85a) by adapting the predistortion parameters of a nonlinear system model so that the deviation between the sensor signal (70) and a virtual sensor signal (70a) at the output of the sink or the deviation between a quantity derived from the sensor signal (70) and a quantity derived from the virtual sensor signal (70a) is minimized.

31. 31. A method (130) according to claim 30, comprising the step of, after performing the minimization of the second objective function (85a), performing a minimization of a first objective function (80, 80a) according to any one of claims 18 to 30 in order to minimize nonlinear distortion in the output of the nonlinear system model.

32. 32. The method (130) of claim 31, comprising outputting a predistorted signal (40, 40a) after minimizing the second objective function (85a) and the first objective function (80, 80a) and communicating the predistorted signal (40, 40a) to the sink (50).

33. 32. The method (130) of claim 31, comprising adapting the at least one parameter and / or the at least one distortion parameter of the nonlinear system model once, continuously, at time intervals, or when a threshold value of the first objective function (80, 80a) or the second objective function (85a) is exceeded.

34. dividing the input signal (20) into frequency bands by a filter bank before converting the input signal (20) into a number of frequency band dependent input signals; predistorting the frequency band dependent input signal (20); aggregating the number of frequency band dependent distorted signals (40, 40a) to form the overall predistorted signal (40, 40a) at the output of the control unit (30, 30a) before transmitting the overall predistorted signal (40, 40a) to the sink (50); 33. The method (130) of any one of claims 16 to 32, comprising:

35. 35. A computer-readable storage medium having stored thereon instructions that, when executed by a computer coupled to a circuit (100), cause the computer to use the circuit (100) to perform a method (130) according to any one of claims 16 to 34.

Images