KR100366289B1 - Predistorter design and method using stochastic gradient method and indirect learning architecture - Google Patents

Abstract

The present invention uses a statistical gradient approximation method to modify the coefficients of a communication system, and attempts to modify coefficients for a complex structure of a precompensator having a combination of linear and nonlinear distortions, thereby improving system performance and increasing power efficiency. It is about.

According to the present invention, a first inverse linear device for compensating for the linear distortion that occurs when there is a synchronization error in the sampling process of the received signal in the reception filter, ISI (Inter-Symbol Interference); An inverse nonlinear device for compensating for nonlinear amplification occurring in the power amplifier using a phase system and a polynomial type; A second inverse linear device for compensating an ISI signal generated when linear distortion occurs in the filtering process of the pulse shaping filter; And a learning device receiving a signal from the second inverse linear apparatus and updating and feeding back a coefficient of a polynomial type used in the inverse nonlinear apparatus using a statistical gradient approximation method and an indirect learning method. Is provided.

Description

Predistorter design and method using stochastic gradient method and indirect learning architecture}

The present invention relates to a precompensator and a method for compensating for linear and nonlinear distortion of a signal occurring in a communication system. More specifically, the present invention uses a statistical gradient approximation method to correct coefficients of each system. The present invention relates to a precompensator that improves the performance and power efficiency of a system by attempting coefficient correction for a complex structure of a precompensator having a complex nonlinear distortion.

In general, if the input power of a power amplifier is increased to increase power efficiency in a communication system, the amplifier operates in a saturation region and nonlinear distortion occurs. Such nonlinear distortion extends the frequency band of the output signal or complexly distorts the output signal due to the action of a filter. Since power amplifiers have a limited output power, when they reach the saturation region, they lose linearity, causing AM / AM distortion, which distorts the waveform of the output signal. In addition, AM / PM distortion occurs in which the phase of the carrier changes according to the power level of the signal.

One way to reduce the effects of this nonlinearity is to increase the back-off of the amplifier input power, but this reduces the efficiency of power usage. Equalizers and precompensators are devices that can efficiently compensate for nonlinearities while using power efficiently. The equalizers compensate for nonlinear distortions on the receiving side, and the precompensators reversely distort before nonlinear distortions occur. Way.

In conventional compensation methods, a precompensator such as a method of modifying contents by a LMS (Least Mean Square) algorithm by placing a look up table (LUT) in front of an amplifier having nonlinearity has been frequently used (WG Jeon, KH). Chang, and YS Cho, "An Adaptive Data Predistoerter for Compensation fo Nonlinear Distortion in OFDM Systems," IEEE Transactions on communications.vol. 45, no. 10, pp. 1167-1171, October 1997).

However, the use of such LUT tables requires high-speed updates and the use of compensators such as Viterbi Algorithm (Bo Hu, Tsz-Mei Ko, "A Viterbi Equalizer for Nonlinear Communications Channels," In the case of 2nd Asia-Pacific Conference on Communications-Conference Record, Osaka, Japan, vol. 2 of 2, 1995), there may be a problem of Noise Enhancement Effect. Compensation techniques such as analog predistorter using Squared Error (Mria-Gabrella Di Benendetto, Paolo Mandarini, "An Application of MMSE Predistortion to OFDM Systems," IEEE Transactions on Communications, vol. 44, no. 11, pp. 1417-1420, November 1996.) However, analog predistorters are difficult to implement.

In addition, many recognition techniques have been proposed for modeling and compensating for nonlinearities.

The Volterra model and the Neural Nework method have been studied for the recognition of dual systems and the design of precompensators [N. benvenuto, M. Marchesi, F. Piazza, A. Uncini, Nonlinear satellite Radio Links Equalized Using Blind neural Networks]. However, in the Volterra model, since the number of coefficients increases exponentially as the order increases, the system becomes complicated in the design of the precompensator, and there is a problem that a lot of overload occurs in the operation of the system.

In addition, in the case of neural networks, stability is higher at higher orders, but at lower orders, efficiency is less effective in compensating for nonlinearities. In addition, there is a problem that neural networks require a large number of coefficients.

Accordingly, the present invention is to solve the above problems of the prior art, an object of the present invention is to provide a simpler and more effective device for precompensation of linear and nonlinear distortion for the recognition and compensation of a nonlinear system. .

For this purpose, in the present invention, by using a statistical gradient approximation method to modify the coefficients of each system, by attempting to modify the coefficients for the precompensator of a complex structure in which linear and nonlinear distortions exist in combination, improving the performance and power efficiency of the system Provides a precompensator that improves the performance and approximates a communication system connecting a transmission filter, amplification and channel, and a reception filter in the baseband region of the communication system using a finite impulse response (FIR) filter and a polynomial nonlinear system. Provide a system model.

1 is a schematic configuration diagram of a structure of an actual communication system and a system modeling the same according to an embodiment of the present invention;

2 is a block diagram briefly illustrating an internal configuration of a predistorter according to an embodiment of the present invention;

3 is a schematic diagram showing the structure of a FIR filter applied to a linear system according to an embodiment of the present invention,

4 is a view showing a characteristic curve of a nonlinear function applied to an embodiment of the present invention,

5 is a block diagram schematically showing the configuration of the overall system to which the pre-compensator according to an embodiment of the present invention is applied.

※ Explanation of code about main part of drawing ※

101: pulse shaping filter 102: power amplifier

103: reception filter 104, 106: linear system

105: nonlinear system 201, 203: inverse linear system

202: inverse nonlinear system 301, 302, 303: delay device

304, 305, 306, 307, 308: Multiplier

309, 310, 311, 312: adder

501: Precompensator

502: Communication System 503: Learning System

504: adder

According to the present invention for achieving the above object, a pulse shaping filter for converting an input discrete signal into a continuous signal, a power amplifier for amplifying a signal passing through the pulse shaping filter and a signal passing through the power amplifier is received A precompensator for compensating for distortion in a communication system including a reception filter, the linearity generated when ISI (Inter-Symbol Interfenence) occurs due to a synchronization error during sampling of a received signal in the reception filter. A first inverse linear device for compensating for distortion; An inverse nonlinear device for compensating for nonlinear amplification occurring in the power amplifier using a phase system and a polynomial type; A second inverse linear device for compensating an ISI signal generated when linear distortion occurs in the filtering process of the pulse shaping filter; And a learning device receiving a signal from the second inverse linear apparatus and updating and feeding back a coefficient of a polynomial type used in the inverse nonlinear apparatus using a statistical gradient approximation method and an indirect learning method. Is provided.

In addition, the distortion of the communication system including a pulse shaping filter for converting the input discrete signal into a continuous signal, a power amplifier for amplifying the signal passed through the pulse shaping filter, and a reception filter for receiving the signal passed through the power amplifier A precompensation method for driving a precompensator for compensating for a phenomenon, the method comprising: compensating for a linear distortion occurring when ISI (Inter-Symbol Interference) occurs due to a synchronization error during sampling of a received signal in the reception filter; First step; Compensating for nonlinear amplification occurring in the power amplifier using a phase system and a polynomial type; A third step of compensating an ISI signal generated when there is a linear distortion in the filtering process of the pulse shaping filter; And a fourth step of receiving the signal compensated in the third step and updating and feeding back a coefficient of the polynomial type used in the inverse nonlinear apparatus by using a statistical gradient approximation method and an indirect learning method. A compensation method is provided.

In addition, the distortion of the communication system including a pulse shaping filter for converting the input discrete signal into a continuous signal, a power amplifier for amplifying the signal passed through the pulse shaping filter, and a reception filter for receiving the signal passed through the power amplifier A computer-readable recording medium recording a program capable of executing a precompensation method for driving a precompensator for compensating for a phenomenon, wherein the reception filter has a synchronization error during sampling of a received signal, thereby causing inter-symbols. Interference: a first step of compensating for linear distortion occurring when intersymbol interference occurs; Compensating for nonlinear amplification occurring in the power amplifier using a phase system and a polynomial type; A third step of compensating an ISI signal generated when there is a linear distortion in the filtering process of the pulse shaping filter; And a fourth step of receiving the signal compensated in the third step and updating and feeding back a coefficient of the polynomial type used in the inverse nonlinear apparatus using a statistical gradient approximation method and an indirect learning method. A computer readable recording medium having recorded a program is provided.

Hereinafter, a precompensator, a communication system modeling apparatus using the same, and a method thereof will be described in detail with reference to the accompanying drawings.

1 is a schematic configuration diagram of a structure of an actual communication system and a system modeling the same according to an embodiment of the present invention.

FIG. 1A is a simplified block diagram of a conventional real communication system, which is composed of a pulse shaping filter 101, a power amplifier 102, and a reception filter 103.

First, the input signal x [n] is composed of a binary signal as a discrete signal. The pulse shaping filter 101 performs a function of converting an input discrete signal into a continuous signal, thereby providing a signal form that can be transmitted by removing high frequency components of the discrete signal.

Subsequently, the power amplifier 102 amplifies the signal passing through the transmission filter to match the transmission power. In an ideal case, linear amplification should occur according to the amplification rate, but there is a saturation region in the actual system. Nonlinear amplification occurs, resulting in signal distortion.

Subsequently, the reception filter 103 receives a signal that has passed through the power amplifier and the channel, and when a synchronization error occurs in the sampling process of the received signal, ISI (Inter-Symbol Interference) occurs.

Figure 1 (b) is a model of the actual communication system shown in Figure 1 (a) described above, consisting of a first linear system 104, a non-linear system 105 and a second linear system 106. do.

The first linear system 104 is composed of a FIR filter fabricated for the recognition operation of the pulse shaping filter 101, a detailed configuration will be described later.

The first linear system 104 generates an ISI interference signal, which is an input component of the nonlinear amplification system 105. The nonlinear system 105 is part of a system associated with a power amplifier of a communication system, and upon power amplification a nonlinear amplified signal is generated. The nonlinear amplification model used in the present invention is a polynomial type nonlinear system model such as Equation 1 below.

y [n] = ax [n] + a ₃ x [n] ³ + a ₅ x [n] ⁵

Here, the value of x [n] is an input value, y [n] is an output value, and a, a ₃ and a ₅ represent coefficients.

As can be seen from Equation 1, the output value y [n] is affected by the first-order term, third-order term, and error-number term of the input value x [n]. Thus, as the input value increases, the third and fifth component values, rather than the primary components, at the output increase relatively, where a nonlinear signal is generated. The use of the nonlinear model of the polynomial type used in the present invention results in significant coefficient gains compared to the existing Volterra and neural network models. The absence of even order terms in the above nonlinear system model is excluded from consideration because the even order terms do not affect in-band in the real system.

On the other hand, the second linear system 106 has the same FIR structure as the first linear system 104 and was used as a linear system model for the recognition of the reception filter.

FIG. 2 is a block diagram schematically illustrating an internal configuration of a precompensator according to an exemplary embodiment of the present invention, and illustrates a compensation system based on a component sequential compensation scheme. Referring to FIG.

The precompensator provided in one embodiment of the present invention consists of a first inverse linear system 201, an inverse nonlinear system 202, and a second inverse linear system 203.

The first inverse linear system 201 is responsible for the compensation for the linear distortion of the receive filter 103, the structure of the receive filter 103 has a FIR filter structure. Therefore, a system inversely needed to be constructed, which appears in the form of an Infinite Impulse Response (IIR) filter. Thus, the inverse system, i.e. the IIR system, depends on the performance of the system depending on the number of coefficients used. However, using a large number of coefficients shows an improvement in the performance of the system, but increases the processing capacity of the system, resulting in many nodes in the system. Therefore, apply the appropriate number of IIR filter coefficients. In the system proposed in the present invention, nine system coefficients are applied.

The inverse nonlinear system 202 compensates for the nonlinear amplification of the power amplifier 102 as a phase system, a compensator having a polynomial type, and distorts the signal by adjusting the respective nonlinear coefficient values.

The second reverse linear system 203 has the same structure as the first reverse linear system 201 and is an inverse system for compensating for distortion of the pulse shaping filter 101 on the transmitting side. Thus, in the overall operation, the predistortion for the receive filter, the predistortion for the nonlinear system, and the predistortion for the transmission filter are performed in this order.

As a result, inverse distortion for linear and nonlinear systems is performed sequentially before being input into the communication system.

3 is a schematic diagram showing the structure of an FIR filter applied to a linear system according to an embodiment of the present invention, wherein the FIR filter includes a signal delay device (301, 302, 303), a multiplier (304, 305, 306, 307). 308, and adders 309, 310, 311, and 312.

The signal delay devices 301, 302, and 303 are devices for delaying one symbol and are constituted by a flip flop or the like. Each system is connected in series, with the values in the (302) system being one symbol more delayed than the values in the (301) system, and the values in the (303) system being (N-1) than the values in the (302) system. It has a value delayed by a symbol. The delay devices 301, 302, and 303 delay the signal by one clock to adjust the amount of interference of the past signal or the future signal to the input signal, and remove the high frequency components on the receiving side.

The multipliers 304, 305, 306, 307, and 308 multiply the values of the signals stored in the memory by the coefficients, respectively, to determine the degree to which each signal affects the input signal, and determine the respective coefficient values. Adjust the influence of the interference component.

The adders 309, 310, 311, and 312 perform a function of calculating and outputting a sum of a component of each signal value multiplied by the coefficient value.

When one signal is input, the operation of the system is affected by the previously input signal, and the interference between the signals depends on the magnitude of each coefficient.

4 is a view showing a characteristic curve of a nonlinear function applied to an embodiment of the present invention.

In theory, the amplifier should have a constant proportional change in output signal size, depending on the size of the input signal and the amplification ratio of the amplifier. However, real power amplifiers do not meet this requirement. Real-world systems have a degree of difference, but maintain linearity when the input power is low, but as the input power gradually increases, the linear characteristics are lost. As shown in FIG. appear. In general, a linear increase occurs in a small input area, i.e., a large back-off input power, but loses linearity as the input power gradually increases, that is, as the back-off power decreases. The value reaches the saturation region where it stops increasing.

5 is a configuration diagram briefly illustrating a configuration of an overall communication system to which a predistorter is applied according to an embodiment of the present invention.

The overall communication system includes a precompensator 501, a communication system 502 in which adjacent signal interference and nonlinear amplification occurs in signal transmission, a learning system 503 for updating respective coefficients of the precompensator 501, and It consists of an adder 504 that calculates the difference between the predistorter 501 and the learning system 503.

The interior of the precompensator 501 has the components shown in FIG. 2, and the input signal undergoes a predistortion process as described above. The communication system 502 has the components shown in FIG. 1A and linear and nonlinear distortions occur. The learning system 503 has the same structure as the predistorter 501, and performs the predistortion process illustrated in FIG. 2 as the predistorter 501. In this case, the output value calculates a difference value between the output value of the precompensator 501 and the value in the adder 504, and the coefficients of the learning system 503 and the precompensator 501 are used based on the difference value. Update the value.

Looking at the operation of the precompensator 501 in detail, as follows.

An input signal passes through the predistorter 501 and the communication system 502, yielding an output signal y [n]. The output signal y [n] is again the input signal of the learning system 503. The learning system 503 has a structure that is completely consistent with the predistorter 501, so that the first inverse linear system 201, the inverse nonlinear system 202, and the second inverse linear system 203 shown in FIG. ). The output signal of the learning system 503 calculates the difference between the output signal of the predistorter 501 and the adder 504. From the meaning of this difference, the input signal x [n] is passed through the precompensator 501 to calculate y [n] output through the actual communication system 502. In this case, when the precompensator 501 performs the complete compensation, the input signal x [n] and the output signal y [n] have exactly the same values. Thus, when passing through the learning system 503 holding the same system as the precompensator 501, the output value is equal to the output value of the precompensator 501. At this time, the difference value in the adder 504 becomes 0, which is a full transposition compensation. The indirect learning system 503 does not need to directly compare the output value y [n] with the input value x [n], and provides an advantage of designing the predistorter in an indirect manner.

A feature of the present invention is the use of a statistical gradient approximation method for coefficient updating and system recognition of the predistorter. Although the LMS (Least Mean Square) method, which is well defined mathematically, has been used in the conventional method for modifying the coefficient of the compensator for nonlinear distortion, the main problems encountered in high-dimensional pattern recognition are as follows.

In other words, a statistical gradient approximation method is used to secure the disadvantage of estimating a matrix that is difficult or impossible to obtain an inverse matrix. This gradient method obtains a desired weight by repeatedly updating the weight in the direction of minimizing the square error between the target value and the output value. Through a simple first derivative, we can obtain a coefficient correction equation as shown in Equation 2 below.

Where n represents the number of learning repetitions every Is the value of the coefficient, Means the current coefficient value, Is the coefficient value to be applied next, and α is an important constant that determines the width to modify each filter coefficient. The convergence speed can be adjusted according to the size. If the value is too large, it can diverge without convergence.

Another feature of the present invention is the use of polynomial forms in the nonlinear model used in the recognition of nonlinear systems and in the design of precompensators. Using these models, the coefficients needed to design the recognition and precompensators have significant benefits over the Volterra and neural network models.

While the invention has been described above based on the preferred embodiments thereof, these embodiments are intended to illustrate rather than limit the invention. It will be apparent to those skilled in the art that various changes, modifications, or adjustments to the above embodiments can be made without departing from the spirit of the invention. Therefore, the protection scope of the present invention will be limited only by the appended claims, and should be construed as including all such changes, modifications or adjustments.

As described above, according to the present invention, a precompensator using the statistical gradient approximation method, the indirect learning system, the component sequential compensation method, and the polynomial nonlinear system has the following advantages.

First, when using the precompensator of the present invention, it can be applied in various fields such as speakers compared to the conventionally used equalizer,

Second, when the sequential compensation method shown in FIG. 2 is applied to the precompensator 501 shown in FIG. 5 and the statistical gradient approximation method is used, linear and nonlinear distortions connected to transmit-amplify-receive are mixed. Compensation for the entire communication system is possible,

Third, by using a polynomial nonlinear system, significant coefficient gains can be obtained over neural networks and Volterra series models.

Fourth, there is an effect of providing a precompensator structure that is not limited to a specific nonlinear communication model and can be adapted to a general nonlinear system.

Claims (12)

A pulse shaping filter for converting an input discrete signal into a continuous signal; a power amplifier for amplifying a signal passing through the pulse shaping filter; and a reception filter for receiving a signal passing through the power amplifier. In the precompensator to compensate, A first inverse linear device for compensating for a linear distortion caused by ISI (Inter-Symbol Interference) due to a synchronization error during sampling of a received signal in the reception filter; An inverse nonlinear device for compensating for nonlinear amplification occurring in the power amplifier using a phase system and a polynomial type; A second inverse linear device for compensating an ISI signal generated when linear distortion occurs in the filtering process of the pulse shaping filter; And A learning device receiving a signal from the second inverse linear device and updating and feeding back a coefficient of a polynomial type used in the inverse nonlinear device using a statistical gradient approximation method and an indirect learning method; Precompensator, characterized in that consisting of. The method of claim 1, The reverse linear device, A precompensator comprising an IIR filter inverse to the FIR filter. The method of claim 2, The FIR filter, A signal delay unit for delaying a series of input signals by one clock to adjust an amount of interference with respect to a past signal or a future signal to the input signal, and to remove a high frequency component of a receiver; The delayed signals are received from the signal delay unit, and these signals are multiplied by weight, respectively, to determine the degree of influence of each signal on the input signal, and to adjust the respective coefficient values to adjust the influence of the interference component. Multiplier; And An adder for calculating and outputting a sum of each signal delayed from the signal delay unit and a signal multiplied by a coefficient from the multiplier; Precompensator, characterized in that consisting of. The method of claim 1, The polynomial type of the inverse nonlinear device is a precompensator according to the following [Formula 1]. [Equation 1] y [n] = ax [n] + a ₃ x [n] ³ + a ₅ x [n] ⁵ Here, the value of x [n] is an input value, y [n] is an output value, and a, a ₃ and a ₅ represent coefficients. The method according to claim 1 or 4, The learning device, And a coefficient of the polynomial type is updated by calculating a difference value between the output signal of the second inverse linear device and the overlay force. The method of claim 5, And updating the coefficient of the polynomial type is determined by Equation 2 below. [Equation 2] Where n represents the number of learning repetitions every Is the value of the coefficient, Means the current coefficient value, Denotes a coefficient value to be applied next, and α is a constant that determines a width for modifying respective filter coefficients. A pulse shaping filter for converting an input discrete signal into a continuous signal; a power amplifier for amplifying a signal passing through the pulse shaping filter; and a reception filter for receiving a signal passing through the power amplifier. A precompensation method for driving a precompensator for compensating, A first step of compensating for a linear distortion occurring when an ISI occurs due to a synchronization error in a sampling process of a received signal in the reception filter; Compensating for nonlinear amplification occurring in the power amplifier using a phase system and a polynomial type; A third step of compensating an ISI signal generated when there is a linear distortion in the filtering process of the pulse shaping filter; And A fourth step of receiving a signal compensated in the third step and updating and feeding back a coefficient of a polynomial type used in the inverse nonlinear apparatus using a statistical gradient approximation method and an indirect learning method; Pre-compensation method comprising a. The method of claim 7, wherein Compensating for the linear distortion, A first sub-step of delaying a series of input signals by one clock to adjust an amount of interference with respect to a past signal or a future signal to the input signal, and to remove a high frequency component of a receiving side; Receive the signals delayed in the first sub-step and multiply these signals by a weight to determine the degree of influence of each signal on the input signal, and adjust the coefficients to adjust the influence of the interference component. A second sub-step; And A third substep of calculating and outputting a sum of a signal obtained by multiplying each signal delayed in the first substep and a coefficient obtained in the second substep; Pre-compensation method comprising a. The method of claim 7, wherein The polynomial type of the second step is the following equation [3]. [Equation 3] y [n] = ax [n] + a ₃ x [n] ³ + a ₅ x [n] ⁵ Here, the value of x [n] is an input value, y [n] is an output value, and a, a ₃ and a ₅ represent coefficients. The method according to claim 7 or 10, The fourth step, And a coefficient of the polynomial type is updated by calculating a difference value between the output signal of the third step and the value obtained in the third substep. The method of claim 10, The updating of the coefficient of the polynomial type is a transposition compensation method, characterized in that determined by the following equation (4). [Equation 4] Where n represents the number of learning repetitions every Is the value of the coefficient, Means the current coefficient value, Denotes a coefficient value to be applied next, and α is a constant that determines a width for modifying respective filter coefficients. A pulse shaping filter for converting an input discrete signal into a continuous signal; a power amplifier for amplifying a signal passing through the pulse shaping filter; and a reception filter for receiving a signal passing through the power amplifier. A computer readable recording medium having recorded thereon a program capable of executing a precompensation method for driving a precompensator for compensating, comprising: A first step of compensating for a linear distortion occurring when an ISI occurs due to a synchronization error in a sampling process of a received signal in the reception filter; Compensating for nonlinear amplification occurring in the power amplifier using a phase system and a polynomial type; A third step of compensating an ISI signal generated when there is a linear distortion in the filtering process of the pulse shaping filter; And A fourth step of receiving a signal compensated in the third step and updating and feeding back a coefficient of a polynomial type used in the inverse nonlinear apparatus using a statistical gradient approximation method and an indirect learning method; A computer-readable recording medium having recorded thereon a program capable of executing what is included.