Patent Grant
9444412
This application is based upon and claims the benefit of priority of the prior Japanese Patent Application No. 2013-211626, filed on Oct. 9, 2013, the entire contents of which are incorporated herein by reference.
The embodiments discussed herein are related to an apparatus and a method for compensating distortion.
In a wireless communication system, a wireless transmission apparatus for transmitting wireless signals is provided with a power amplifier for amplifying the transmission signals. The power amplifier exhibits a linear characteristic where, in a smaller amplitude region of an input signal, the amplitude of the input signal has a roughly linear relationship with the amplitude of a corresponding output signal (the region may be hereinafter referred to as the “linear region”). On the other hand, the power amplifier exhibits a nonlinear characteristic where, in a larger amplitude region of the input signal, the amplitude of the input signal has a nonlinear relationship with the amplitude of the output signal (the region may be hereinafter referred to as the “nonlinear region”). In view of the efficient use of the power amplifier, it is often the case that the power amplifier is operated not only for the linear region but also for the nonlinear region.
However, nonlinear distortion of a transmission signal due to the nonlinear amplification characteristic of the power amplifier may reduce quality in wireless communication, for example, causing power leakage from an intended frequency range into adjacent frequency ranges. In view of this, some wireless transmission apparatuses are designed to carry out pre-distortion compensation as a means of linearizing the nonlinear amplification characteristic of the power amplifier. For the pre-distortion compensation, such a wireless transmission apparatus is provided with a pre-distorter functioning as a linearizer.
The pre-distorter adds nonlinear distortion having an inverse characteristic to the nonlinear characteristic of the power amplifier to a transmission signal prior to the transmission signal being input to the power amplifier. By feeding the transmission signal with the inverse-characteristic nonlinear distortion added thereto to the power amplifier, it is possible to suppress nonlinear distortion of the transmission signal after being amplified, which nonlinear distortion is generated with respect to the transmission signal prior to the transmission signal being input to the pre-distorter. The pre-distorter nonlinearly changes the amplitude of the transmission signal to be fed to the power amplifier by, for example, applying a compensation coefficient according to the amplitude of the transmission signal to the transmission signal (for example, multiplying the transmission signal by the compensation coefficient). Such compensation coefficients each corresponding to the magnitude of the amplitude of the transmission signal may be stored, for example, in a lookup table (LUT).
It is not easy to accurately estimate in advance the inverse characteristic to the nonlinear characteristic of the power amplifier. Therefore, in the pre-distortion compensation, the compensation coefficients used by the pre-distorter may be adaptively updated as the power amplifier is being operated. For example, the wireless transmission apparatus feeds back output signals of the power amplifier, and estimates appropriate compensation coefficients based on the feedback signals and corresponding transmission signals prior to the transmission signals being input to the power amplifier, which compensation coefficients are then set in the pre-distorter.
Note that a distortion compensation apparatus has been proposed which corrects a compensation coefficient in such a manner that the power of a transmission signal subjected to distortion compensation by a pre-distorter does not exceed the dynamic range of a digital analog converter (DAC). The distortion compensation apparatus is designed to calculate the compensation coefficient based on a feedback signal and then compare, before storing the calculated compensation coefficient in a memory, an estimate of the power of the transmission signal to be output from the pre-distorter in the case of using the compensation coefficient against predetermined upper limit power. If the estimate exceeds the upper limit power, the distortion compensation apparatus corrects the calculated compensation coefficient in such a manner as to reduce the magnitude thereof while maintaining its phase before storing it in the memory.
In addition, a digital pre-distortion system has been proposed in which an input signal is clipped before being fed into a pre-distorter so as to reduce the peak-to-average power ratio of the signal.
In addition, a distortion compensation circuit has been proposed which limits the amplitude of transmission signals before feeding the transmission signals into a pre-distorter so that the amplitude of the transmission signals subjected to distortion compensation does not exceed a saturation level of a power amplifier. The distortion compensation circuit calculates a threshold based on compensation coefficients stored in a memory, and determines whether to limit the amplitude of each of the transmission signals by comparing the calculated threshold and the power of the transmission signal.
Further, a distortion compensation apparatus has been proposed in which an auto gain control (AGC) loop including an attenuation circuit is provided in a digital pre-distorter (DPD) loop including a distortion compensation circuit, an amplifier circuit, and a distortion detection circuit. To prevent operational instability caused by the double-loop structure including the DPD loop and the AGC loop, the distortion compensation apparatus normalizes compensation coefficients when updating the compensation coefficients in such a manner that the gain of the DPD loop becomes 1.
Japanese Laid-open Patent Publication No. 2001-251148
Japanese Laid-open Patent Publication No. 2007-37110
Japanese Laid-open Patent Publication No. 2009-290384
Japanese Laid-open Patent Publication No. 2010-103834
In general, an amplification factor in the nonlinear region of the power amplifier is smaller than an ideal amplification factor for the case where the amplification characteristic is linearized. Therefore, when the pre-distorter is being trained, the maximum amplitude of feedback signals is smaller than that of corresponding input signals in an early stage of the training where compensation coefficients corresponding to the nonlinear region have yet to become sufficiently large. That is, the dynamic range of the feedback signals is smaller than that of the input signals.
A distortion compensation apparatus learns compensation coefficients based on such feedback signals. Therefore, in the early stage of the training, not much learning takes place with respect to the higher amplitude side of the dynamic range of input signals, thus leaving an unlearned region. On the other hand, as the training progresses, compensation coefficients become larger in order from the lower amplitude side of the nonlinear region, thereby gradually increasing the maximum amplitude of the feedback signals and thus narrowing the unlearned region. Eventually, the dynamic range of the feedback signals matches that of the input signals, leaving no unlearned region, and thus the learning of compensation coefficients with respect to the entire dynamic range of the input signals is completed.
However, if the learning of compensation coefficients proceeds in the above-described manner, the unlearned region exists until the middle of the training, and therefore the training needs a long period of time to obtain highly accurate compensation coefficients for the entire range of the input signals.
According to one embodiment, there is provided a distortion compensation apparatus for compensating nonlinear distortion of a power amplifier. The distortion compensation apparatus includes a pre-distorter configured to add distortion according to compensation coefficients to individual input signals prior to the input signals being input to the power amplifier; a gain control unit configured to apply gain control to individual feedback signals fed back from the power amplifier according to a maximum level of the feedback signals within a time frame; and a learning unit configured to update the compensation coefficients used by the pre-distorter, using the feedback signals subjected to the gain control.
The object and advantages of the invention will be realized and attained by means of the elements and combinations particularly pointed out in the claims.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are not restrictive of the invention.
Embodiments are described below with reference to the accompanying drawings.
(a) First Embodiment
The distortion compensation apparatus 10 includes a pre-distorter 11, a gain control unit 12, and a learning unit 13. The pre-distorter 11 adds distortion according to a compensation coefficient to each input signal prior to the input signal being input to the power amplifier 2. The distortion added by the pre-distorter 11 preferably has an inverse characteristic to the nonlinear characteristic of the power amplifier 2 so as to cancel out the nonlinear distortion generated by the power amplifier 2. Compensation coefficients used by the pre-distorter 11 are calculated by training, as described later. For example, the pre-distorter 11 acquires a compensation coefficient corresponding to the amplitude of each input signal from a lookup table in which multiple amplitude levels are associated with multiple compensation coefficients, and applies the acquired compensation coefficient to the input signal (for example, multiplies the input signal by the compensation coefficient).
The gain control unit 12 acquires feedback signals fed back from the power amplifier 2, and applies gain control to each of the feedback signals and provides the resultant signals to the learning unit 13. A gain applied to each of the feedback signals is adjusted according to a maximum level of the feedback signals within a time frame. In order to make the gain variable, the gain control unit 12 may include a variable amplifier or a variable attenuator. For example, the gain control unit 12 detects a maximum level of feedback signals within a time frame and a maximum level of input signals prior to the input signals being input to the pre-distorter 11 within the time frame. Subsequently, the gain control unit 12 determines a gain to be applied to the feedback signals within the time frame, based on the maximum level of the feedback signals and the maximum level of the input signals.
In an early stage of training to be described later, the maximum level of the feedback signals is smaller than the maximum level of the input signals. Therefore, for example, the gain control unit 12 increases the amplitude of each feedback signal by making the gain larger than 1 so that the maximum level of the feedback signals becomes close to the maximum level of the input signals. Preferably, the gain control unit 12 determines a ratio of the maximum level of the input signals to that of the feedback signals as the gain, thereby matching the maximum levels of the feedback signals and the input signals. Note that, as an index indicating the “levels” of the feedback signals and the input signals, amplitude, power, or the like may be used.
During the training of the pre-distorter 11, the learning unit 13 repeatedly updates compensation coefficients used by the pre-distorter 11, using feedback signals subjected to gain control by the gain control unit 12. If a longer time is spent on the training, the accuracy of the compensation coefficients becomes higher. When the learning unit 13 updates compensation coefficients during the training, the pre-distorter 11 carries out distortion compensation using the updated compensation coefficients, and the power amplifier 2 amplifies input signals to which distortion has been added according to the updated compensation coefficients. In the case of continuing the training, the learning unit 13 further updates the compensation coefficients using feedback signals obtained after the last update. At this point, the gain control unit 12 may apply, to the feedback signals, a gain different from a previous one (i.e., a gain for the previous time frame).
As architecture for calculating compensation coefficients, for example, an indirect learning architecture is used. In the indirect learning architecture, the learning unit 13 tries to compensate distortion of each feedback signal subjected to gain control by the gain control unit 12 in such a manner that the amplitude or the phase of the feedback signal comes close to that of a corresponding input signal having passed through the pre-distorter 11. That is, the learning unit 13 calculates errors between individual signals generated by applying current compensation coefficients (currently used by the pre-distorter 11) to the feedback signals and corresponding input signals having passed through the pre-distorter 11, and corrects the compensation coefficients based on the errors. In this case, it may be said that the learning unit 13 is provided with a post-distorter. The leaning unit 13 copies, to the pre-distorter 11, the compensation coefficients learned with the use of the feedback signals.
Assuming here that no gain control by the gain control unit 12 is applied, the maximum amplitude of the feedback signals is smaller than that of the corresponding input signals input to the pre-distorter 11 in an early stage of the training. This is because an amplification factor in the nonlinear region of the power amplifier 2 is smaller than an ideal amplification factor for the case where the amplification characteristic is linearized. That is, in the early stage of the training, the dynamic range of the feedback signals is smaller than that of the input signals input to the pre-distorter 11.
In that case, in the early stage of the training, the learning unit 13 learns compensation coefficients not for the entire dynamic range of the input signals but for only a part of the dynamic range (a part corresponding to the dynamic range of the feedback signals). An unlearned region of the dynamic range of the input signals, from which compensation coefficients have not been learned, sequentially becomes smaller as the training continues. This is because, with the progress of the learning of compensation coefficients for part of the nonlinear region (i.e., as the compensation coefficients become larger), the maximum amplitude of the feedback signals becomes sequentially larger. Note however that a long time may be needed to obtain sufficiently accurate compensation coefficients because the unlearned region exists until the middle of the training.
On the other hand, in the case where gain control is applied to the feedback signals by the gain control unit 12, the maximum amplitude of the feedback signals comes close to that of the input signals input to the pre-distorter 11 (preferably, the two match each other) even in the early stage of the training. In this case, the learning unit 13 is able to learn compensation coefficients from a larger part of the dynamic range of the input signals (preferably, from the entire dynamic range of the input signals) compared to the case where no gain control is applied. Because the unlearned region is reduced (preferably, eliminated) from the early stage of the training, the learning of compensation coefficients is made more efficient, thereby shortening the training time.
Note that in the case where gain control by the gain control unit 12 has been applied, the learning unit 13 overestimates the amplification factor of the power amplifier 2 in the early stage of the training. As a result, compensation coefficients calculated by the learning unit 13 become smaller than ideal compensation coefficients. The underestimation of the compensation coefficients associated with the gain control for the feedback signals is sequentially corrected by continuing the training. Note however that because the unlearned region is smaller compared to the case where no gain control for the feedback signals is provided, it takes a shorter time to obtain sufficiently accurate compensation coefficients for the entire dynamic range of the input signals.
(b) Second Embodiment
Although the second embodiment mainly describes a transmission process of the wireless transmission apparatus 100, a wireless communication apparatus equivalent to the wireless transmission apparatus 100 may further carry out a reception process, or a wireless communication apparatus equivalent to the wireless reception apparatus 200 may further carry out the transmission process. In addition, the wireless communication system may be a one-to-many connection system in which a wireless base station communicates with multiple wireless terminals. In that case, for example, the wireless transmission apparatus 100 may be the wireless base station and the wireless reception apparatus 200 may be one of the wireless terminals. On the other hand, for example, the wireless transmission apparatus 100 may be one of the wireless terminals and the wireless reception apparatus 200 may be the wireless base station. Alternatively, at least one of the wireless transmission apparatus 100 and the wireless reception apparatus 200 may be a wireless relay apparatus.
The DPD 111 is a pre-distorter for carrying out digital distortion compensation for input signals to be input to the power amplifier 112. The DPD 111 has a set of compensation coefficients (hereinafter, simply referred to as a “compensation coefficient set”) used for the distortion compensation. For example, the DPD 111 selects a compensation coefficient according to the amplitude of an input signal and multiplies the input signal by the selected compensation coefficient. The compensation coefficient set is stored, for example, in a lookup table where a compensation coefficient to be used is determined based on the amplitude of an input signal.
The compensation coefficient set preferably corresponds to an inverse characteristic to a nonlinear characteristic of the power amplifier 112, and is used to cancel out nonlinear distortion generated by the power amplifier 112. The compensation coefficient set is determined by training the DPD 111. In the training, N sample signals (N is an integer equal to or greater than 2) having a predetermined length are input to the DPD 111 with respect to each iteration for learning a compensation coefficient set. During the training, the DPD 111 carries out distortion compensation on the sample signals, as in the case of wirelessly communicating with the wireless reception apparatus 200.
For each learning iteration using N sample signals described above, a compensation coefficient set calculated by the training DPD 125 to be described later is copied to the DPD 111. Then, the DPD 111 carries out distortion compensation on the next N sample signals using the updated compensation coefficient set. In this manner, learning with the use of N sample signals is repeated during the training and, therefore, the compensation coefficient set is repeatedly updated. As the number of learning iterations increases, the accuracy of the compensation coefficients becomes higher. Note that, in the first learning iteration, the DPD 111 may use initial values of the compensation coefficients. The initial values of the compensation coefficients may be values having no correction for the sample signals.
Here, the number N of sample signals used in one learning iteration may be set, for example, to correspond to one wireless frame (for example, 10 ms). As the number of sample signals used in one learning iteration increases, the accuracy of the learning becomes higher. The sample signals may be randomly generated, or may be prepared in advance for the training. In the former case, individual sets of sample signals used for multiple learning iterations may be different from each other.
The power amplifier 112 amplifies input signals on which distortion compensation has been implemented by the DPD 111. A DAC may be provided between the DPD 111 and the power amplifier 112. The power amplifier 112 may be an HPA. When the wireless transmission apparatus 100 wirelessly communicates with the wireless reception apparatus 200, transmission signals amplified by the power amplifier 112 are output from the antenna 101. During the training of the DPD 111, the power amplifier 112 amplifies sample signals on which distortion compensation has been implemented by the DPD 111. Here, the power amplifier 112 has such a nonlinear amplification characteristic that the amplitude of input signals has a nonlinear relationship with that of output signals. Therefore, if no distortion compensation is implemented by the DPD 111, the amplified output signals may have nonlinear distortion.
During the training, the attenuator 121 attenuates signals output from the power amplifier 112 and, then, provides the attenuated output signals as feedback signals. An analog digital converter (ADC) may be provided posterior to the attenuator 121. The attenuator 121 multiplies the amplitude of the signals output from the power amplifier 112 by a real number. The attenuation rate of the attenuator 121 may be preliminarily set to a fixed value. For example, the magnification factor of the amplitude used by the attenuator 121 may be the reciprocal of an ideal amplification factor obtained when the amplification characteristic of the power amplifier 112 is linearized.
In each learning iteration, the buffer 122 temporarily stores therein feedback signals (corresponding to the N sample signals) provided by the attenuator 121. As the buffer 122, a volatile or non-volatile semiconductor memory may be used. The buffer 122 delays delivery of the feedback signals to the AGC unit 123 while the gain calculating unit 129 calculates a feedback gain to be used in the current learning iteration, as to be described later.
Once the gain calculating unit 129 calculates the feedback gain to be used in the current learning iteration, the AGC unit 123 applies gain control to the feedback signals stored in the buffer 122 according to the calculated feedback gain. For example, the AGC unit 123 multiplies the individual feedback signals by the calculated feedback gain, thereby achieving multiplication of the amplitude of the feedback signals by a real number. The same feedback gain is applied to the multiple feedback signals used in one learning iteration. In a different learning iteration, a different feedback gain may be applied. In an early stage of the training, the feedback gain is generally larger than 1 and, therefore, the amplitude of the feedback signals having passed through the AGC unit 123 becomes larger than that of the feedback signals before entering the AGC unit 123.
In each learning iteration, the buffer 124 temporarily stores therein input signals (corresponding to the N sample signals) which have undergone distortion compensation of the DPD 111 but have yet to be input to the power amplifier 112. As the buffer 124, a volatile or non-volatile semiconductor memory may be used. The buffer 124 delays delivery of the input signals to the subtractor 126 while the gain calculating unit 129 calculates the feedback gain to be used in the current learning iteration.
During the training, the training DPD 125 learns a compensation coefficient set by carrying out distortion which is the same as the pre-distortion made by the DPD 111. In view of this, it may be said that the training DPD 125 operates as a post-distorter. Specifically, the training DPD 125 has the same compensation coefficient set as the DPD 111 at the start of the first learning iteration using the N sample signals. The compensation coefficient set is stored, for example, in a lookup table where compensation coefficients are determined based on the amplitude of the individual signals.
In a single learning iteration, the training DPD 125 carries out distortion compensation on the feedback signals acquired from the AGC unit 123 and, then, outputs the feedback signals with distortion compensated to the subtractor 126. For example, the training DPD 125 selects a compensation coefficient according to the amplitude of each of the feedback signals acquired from the AGC unit 123, and multiplies the feedback signal by the selected compensation coefficient. Subsequently, the training DPD 125 acquires, from the subtractor 126, information indicating errors of the feedback signals with distortion compensated. Then, the training DPD 125 updates its owned compensation coefficient set based on the errors.
For example, the training DPD 125 calculates the root mean square of the errors, and updates its owned compensation coefficient set in such a manner as to minimize the overall error based on the least mean square (LMS) algorithm. Then, when a single learning iteration using the N sample signals is finished, the compensation coefficient set of the training DPD 125 is copied to the DPD 111. During the training, such learning of a compensation coefficient set is repeated. The iteration count may be preliminarily set to a fixed number, or may be adaptively determined according to the errors calculated by the subtractor 126. Note that the above-described learning method is sometimes referred to as the indirect learning architecture.
The subtractor 126 acquires the feedback signals with distortion compensated from the training DPD 125, and also acquires input signals corresponding to the feedback signals from the buffer 124. The subtractor 126 compares the input signals and the feedback signals to calculate their differences, and notifies the training DPD 125 of the calculated differences as linearization errors.
In each learning iteration, the maximum value detecting unit 127 acquires input signals (sample signals) to be input to the DPD 111, and detects the maximum amplitude of the input signals. The maximum amplitude is a maximum value of the amplitude of the N input signals, and one maximum amplitude value is obtained for each learning iteration.
In each learning iteration, the maximum value detecting unit 128 acquires feedback signals from the attenuator 121, and detects the maximum amplitude of the feedback signals. The maximum amplitude is a maximum value of the amplitude of the N feedback signals, and one maximum amplitude value is obtained for each learning iteration.
The gain calculating unit 129 calculates a feedback gain to be used by the AGC unit 123, based on the maximum amplitude of the input signals detected by the maximum value detecting unit 127 and the maximum amplitude of the feedback signals detected by the maximum value detecting unit 128. The feedback gain is calculated for each learning iteration. According to the second embodiment, the feedback gain is obtained by dividing the maximum amplitude of the input signals by the maximum amplitude of the feedback signals. With this, the maximum amplitude of the feedback signals subjected to gain control matches the maximum amplitude of the input signals. As to be described later, the feedback gain calculated by the gain calculating unit 129 becomes larger than 1 because, in an early stage of the training, the maximum amplitude of the feedback signals is smaller than that of the input signals.
Note that the attenuator 121, the AGC unit 123, the training DPD 125, the subtractor 126, the maximum value detecting units 127 and 128, and the gain calculating unit 129 may be caused to operate only during the training, and may be stopped from operating when the wireless transmission apparatus 100 wirelessly communicates with the wireless reception apparatus 200.
Next described are characteristics of the DPD 111 and the power amplifier 112 and the role of the AGC unit 123.
The DPD 111 preferably produces distortion on input signals in such a manner as to cancel out the nonlinear amplification characteristic of the power amplifier 112, allowing the amplification characteristic of the power amplifier 112 to be linearized. The gain characteristic of the DPD 111 is substantially linear in a region where the amplitude of input signals is small (a linear region), as in the case of the amplification characteristic of the power amplifier 112. In the linear region, the amplitude of output signals does not have to be changed much from the amplitude of input signals. On the other hand, in a region where the amplitude of input signals is large, the gain characteristic of the DPD 111 is nonlinear according to the nonlinear amplification factor of the power amplifier 112 (a nonlinear region). Contrary to the case of the power amplifier 112, the gain becomes larger in the nonlinear region than in the linear region, and the magnification factor of the amplitude increases more as the amplitude of the input signals becomes larger.
Here, the maximum amplitude of the feedback signals in the Mth learning iteration is denoted as y_max[M], and the maximum amplitude of the input signals input to the DPD 111 is denoted as x_max. In the first learning iteration, the training DPD 125 updates compensation coefficients for a region where the amplitude is equal to and less than y_max[1], which is the dynamic range of the feedback signals. On the other hand, the training DPD 125 does not update compensation coefficients for a region where the amplitude is larger than y_max[1] and equal to and less than x_max. The region for which no compensation coefficients are updated may be referred to as the unlearned region. Since the compensation coefficients calculated by the training DPD 125 are copied to the DPD 111, the region where the amplitude is larger than y_max[1] and equal to and less than x_max becomes an unlearned region also for the DPD 111.
In the second iteration for learning compensation coefficients, distortion compensation is carried out on input signals using the compensation coefficient set including the unlearned region. Here, several methods may be considered for processing input signals belonging to the unlearned region. For example, the DPD 111 converts the amplitude of input signals belonging to the unlearned region to a fixed value of the maximum amplitude of output signals in the learned region (for example, the amplitude of an output signal for the case where the amplitude of an input signal is y_max[1]). In addition, for example, assuming that the amplification characteristic of the unlearned region is linear, the DPD 111 increases the amplitude of input signals belonging to the unlearned region by the maximum magnification of the learned region (for example, by the magnification for the case where the amplitude of an input signal is y_max[1]).
Because the learned region includes part of the nonlinear region, compensation coefficients of the DPD 111 become large for at least part of the nonlinear region. As a result, the maximum amplitude of the feedback signals in the second learning iteration becomes larger than that in the first learning iteration. That is, y_max[1]<y_max[2], and the unlearned region of the DPD 111 becomes smaller. In a similar fashion, by repeating the learning of compensation coefficients, the maximum amplitude of the feedback signals becomes larger in a stepwise manner like y_max[2]<y_max[3]<y_max[4] while the unlearned region of the DPD 111 becomes smaller in a stepwise manner. Eventually, the maximum amplitude of the feedback signals substantially matches x_max and there is no more unlearned region for the DPD 111. Note however that because compensation coefficients for higher amplitude are not learned until the middle of the training, the training takes a long time to increase the accuracy of the compensation coefficients.
Herewith, the maximum amplitude of the feedback signals acquired by the training DPD 125 in the first learning iteration is corrected from y_max[1] to x_max. As a result, the training DPD 125 is able to update compensation coefficients for the entire dynamic range of the input signals where the amplitude is equal to and less than x_max, and thus no unlearned region is present. In a similar fashion, the compensation coefficients are updated for the entire dynamic range of the input signals from the second learning iteration onward.
Note that the amplitude of the feedback signals being increased by the AGC unit 123 means that the training DPD 125 overestimates the amplification factor of the power amplifier 112. Therefore, in an early stage of the training, the training DPD 125 calculates such compensation coefficients smaller than ideal compensation coefficients that the amplitude of the input signals with distortion compensated is kept small. The difference between the distortion compensation coefficients is resolved in a stepwise manner by repeating the compensation coefficient learning. Note however that, because no unlearned region is present from the first learning iteration, it is possible to reduce the training time needed to sufficiently improve the accuracy of the compensation coefficients compared to the case in
Next described are flows of signals in the wireless transmission apparatus 100.
The maximum value detecting unit 127 acquires the signals x(n) to be input to the DPD 111, and detects the maximum amplitude x_max of the signals x(1) to x(N) which are N input signals. The maximum value detecting unit 128 acquires the signals y(n)/G from the attenuator 121, and detects the maximum amplitude y_max of the signals y(1)/G to y(N)/G which are N feedback signals. The gain calculating unit 129 calculates a feedback gain FG obtained by dividing the maximum amplitude x_max by the maximum amplitude y_max (FG=x_max÷y_max). The AGC unit 123 applies the feedback gain FG to the signals y(n)/G held in the buffer 122 to thereby provide signals y(n)·FG/G to the training DPD 125.
The training DPD 125 applies compensation coefficients to the signals y(n)·FG/G to thereby output signals z^(n). When the input of the N sample signals is started, the training DPD 125 has a compensation coefficient set A which is the same as a compensation coefficient set held by the DPD 111. The subtractor 126 calculates errors E by subtracting the signals z^(n) output from the training DPD 125 from the signals z(n) held in the buffer 124. The training DPD 125 uses the LMS algorithm with the errors E associated with the N sample signals to thereby update the compensation coefficient set A. Subsequently, the training DPD 125 copies the compensation coefficient set A to the DPD 111.
Next described is a procedure for training the DPD 111.
(S1) The wireless transmission apparatus 100 starts inputting N sample signals to the DPD 111. The sample signals are, for example, randomly generated signals corresponding to one wireless frame. The DPD 111 carries out distortion compensation on the N input signals which are sample signals. The power amplifier 112 amplifies the N input signals with distortion compensated. The attenuator 121 attenuates outputs of the power amplifier 112 at a predetermined attenuation rate to thereby obtain N feedback signals.
(S2) The maximum value detecting unit 127 monitors input signals entering the DPD 111, and then detects a maximum amplitude value (the maximum amplitude x_max) of the N input signals. The maximum amplitude x_max is determined after all the N input signals are supplied to the DPD 111.
(S3) The maximum value detecting unit 128 monitors feedback signals output from the attenuator 121, and then detects a maximum amplitude value (the maximum amplitude y_max) of the N feedback signals. The maximum amplitude y_max is determined after all the N feedback signals are generated.
(S4) The gain calculating unit 129 determines the feedback gain FG to be used in the current learning iteration, based on the maximum amplitude x_max detected in step S2 and the maximum amplitude y_max detected in step S3. The feedback gain FG is obtained by dividing the maximum amplitude x_max by the maximum amplitude y_max (FG=x_max y_max). Note that at the time when the feedback gain FG is determined, the N feedback signals are stored in the buffer 122 and the N input signals are stored in the buffer 124.
(S5) Based on the feedback gain FG determined in step S4, the AGC unit 123 applies gain control to the N feedback signals stored in the buffer 122. That is, the AGC unit 123 multiplies the amplitude of each of the feedback signals by the feedback gain FG. With this, the original dynamic range of the feedback signals (the amplitude range up to y_max) is expanded to the dynamic range of the input signals before the distortion compensation (the amplitude range up to x_max).
(S6) The training DPD 125 updates the compensation coefficient set A using the N feedback signals subjected to the gain control in step S5. That is, the training DPD 125 carries out distortion compensation on the feedback signals using its owned compensation coefficient set A (the same as that of the DPD 111). The subtractor 126 calculates errors between the feedback signals with distortion compensated and the input signals with distortion compensated, which input signals are stored in the buffer 124. The training DPD 125 applies a predetermined algorithm to the errors associated with the N feedback signals to thereby calculate such a compensation coefficient set A as to minimize the errors.
(S7) The training DPD 125 copies the compensation coefficient set A updated in step S6 to the DPD 111. With this, one iteration for learning the compensation coefficient set A is finished. In the following learning iteration, the updated compensation coefficient set A is used.
(S8) The wireless transmission apparatus 100 determines whether to finish the iterations for learning the compensation coefficient set A, that is, whether to finish the training of the DPD 111. The iteration number for learning the compensation coefficient set A may be preliminarily set to a fixed value, or may be dynamically determined according to the convergence condition of the compensation coefficient set A. In the case of repeating the learning of the compensation coefficient set A, the wireless transmission apparatus 100 proceeds to step S1. If not, the wireless transmission apparatus 100 ends the training of the DPD 111.
The ACLR becomes larger as the nonlinear distortion caused when the power amplifier 112 amplifies input signals is larger, and the ACLR becomes smaller as the accuracy of the distortion compensation achieved by the DPD 111 is higher (i.e., as the accuracy of the learned compensation coefficients is higher). An increase in the iteration number for learning compensation coefficients provides higher accuracy of the compensation coefficients (i.e., provides compensation coefficients closer to ideal ones), thus decreasing the ACLR.
Assume here that a target ACLR is −5.7E+01 dB, and that the learning of compensation coefficients is repeated until the ACLR becomes at least smaller than the target ACLR. According to the example of
According to the wireless transmission apparatus 100 of the second embodiment, the maximum amplitude of the feedback signals used in the learning of compensation coefficients matches the maximum amplitude of the input signals even in an early stage of the training. Therefore, it is possible to resolve the problem of an unlearned region from the early stage of the training, which allows the learning of compensation coefficients for the entire dynamic range of the input signals. With this, the compensation coefficients converge faster, thus resulting in a shorter training time. In addition, it is possible to efficiently improve the accuracy of the compensation coefficients within a limited training time. This reduces power leakage into adjacent channels, contributing to an improvement in communication quality of the wireless transmission apparatus 100.
According to one aspect, it is possible to reduce the training time of the pre-distorter.
All examples and conditional language provided herein are intended for the pedagogical purposes of aiding the reader in understanding the invention and the concepts contributed by the inventor to further the art, and are not to be construed as limitations to such specifically recited examples and conditions, nor does the organization of such examples in the specification relate to a showing of the superiority and inferiority of the invention. Although one or more embodiments of the present invention have been described in detail, it should be understood that various changes, substitutions, and alterations could be made hereto without departing from the spirit and scope of the invention.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2013-211626 | Oct 2013 | JP | national |
| Number | Name | Date | Kind |
|---|---|---|---|
| 20080068080 | Miyatani | Mar 2008 | A1 |
| 20080197925 | Furuta | Aug 2008 | A1 |
| 20090115513 | Hongo | May 2009 | A1 |
| 20120256687 | Davies | Oct 2012 | A1 |
| 20130040587 | Ishikawa | Feb 2013 | A1 |
| 20140254716 | Zhou | Sep 2014 | A1 |
| Number | Date | Country |
|---|---|---|
| 2001-251148 | Sep 2001 | JP |
| 2007-37110 | Feb 2007 | JP |
| 2009-290384 | Dec 2009 | JP |
| 2010-103834 | May 2010 | JP |
| Entry |
|---|
| Lei Ding et al., “A Robust Digital Baseband Predistorter Constructed Using Memory Polynomials”, IEEE Transactions on Communications, vol. 52, No. 1, Jan. 2004, pp. 159-165. |
| Number | Date | Country | |
|---|---|---|---|
| 20150097618 A1 | Apr 2015 | US |
