In many technical fields, linearization of a nonlinear element is used to compensate for unwanted effects caused by the nonlinear behaviour of the nonlinear element. One possibility to linearize a nonlinear element is to predistort the signal input into the nonlinear element to ensure that the output signal of the nonlinear element is, in the ideal case, linearly related to the input signal of the predistorter.
In general, such predistorter is a highly complex nonlinear system which incorporates memory. An important step in predistortion is the identification of the nonlinear element based on input-output measurements. Once the nonlinear element has been identified, it may prove possible to calculate appropriate predistorter parameters for linearization of the system containing the predistorter and the nonlinear element.
The technique of predistortion is widely used in modern wireline and wireless communication systems. Typically, communication systems employ a power amplifier with high output power requirements. Such power amplifiers are often driven in the nonlinear region to obtain the highest possible efficiency. This leads in general to spectral regrowth and intermodulation distortion in the signal band. Basically, there are two approaches to minimize these unwanted effects. The first one is the employment of an oversized power transistor which is purely driven in the linear range even for maximum output power requirements. This approach is highly cost intensive during production and also during operation because of the high price of such transistor and its low signal to DC efficiency. The second approach, signal predistortion, allows to create low price devices fulfilling a given spectral mask for the transmission signal even though the power amplifier is driven in the nonlinear region.
Identification of the nonlinear power amplifier or, more general, the nonlinear element is usually accomplished at a sufficiently high sampling rate to cover the information in the out-off-band region caused by the nonlinearity. Usually, the sampling frequency for identification of the nonlinear element is at least twice the bandwidth of the output signal, i.e. 2P-times higher than the signal bandwidth at the input of the nonlinear element, where the factor P denotes the highest odd-order nonlinearity of the nonlinear element. Because the nonlinear behaviour of the nonlinear element is generally unknown before identification, high sampling rates of a few hundreds of Msamples/s are used in communication applications.
According to a generalized sampling theorem, it is possible to uniquely identify a nonlinear element on a sampling frequency which is twice the input signal bandwidth or, in other words, which can satisfy the Nyquist theorem for the input signal of the nonlinear element.
The following presents a simplified summary of the invention in order to provide a basic understanding of some aspects of the invention. This summary is not an extensive overview of the invention, and is neither intended to identify key or critical elements of the invention nor to delineate the scope of the invention. Rather, the purpose of the summary is to present some concepts of the invention in a simplified form as a prelude to the more detailed description that is presented later.
In one embodiment, the invention relates to a circuit that includes an adaptive predistorter to predistort an input signal according to a predistortion control parameter. A nonlinear element is coupled to an output of the predistorter, the output having a bandwidth. The circuit also includes an identification unit to identify a parameter representing the nonlinearity of the nonlinear element and to sample the output of the nonlinear element at an identification sampling rate lower than twice the bandwidth at the output of the nonlinear element. In addition, the circuit includes a translation unit to translate the parameter into the predistortion control parameter.
The following description and annexed drawings set forth in detail certain illustrative aspects and implementations of the invention. These are indicative of but a few of the various ways in which the principles of the invention may be employed.
Aspects of the invention are made more evident in the following detailed description of some embodiments when read in conjunction with the attached drawing figures, wherein:
In the following description and claims, the term Volterra kernel is used. This term has a well defined meaning in nonlinear system theory. The concept of describing nonlinear systems in terms of Volterra kernels will not be explained in detail herein. For an intuitive understanding of the concept of Volterra kernels, it is to be noted that Volterra kernels are used to describe a nonlinear system in a similar way as the impulse response is used to describe a linear system. In linear system theory, the output signal of a linear system is the convolution of the input signal with the impulse response. Analogously, the output signal of a nonlinear system is the multi-dimensional convolution of the input signal with a series expansion of Volterra kernels. In fact, the first order Volterra kernel is identical to the impulse response of a linear system. As the concept of describing a system by an impulse response is limited to linear systems, the Volterra kernel representation may be intuitively understood as a generalization of the impulse response concept to nonlinear systems.
The predistorter 1 acts on the digital input signal u(n) by a nonlinear operator which is in one embodiment the inverse of the nonlinearity of the power amplifier 3 to obtain an overall linear response. Because the nonlinear behavior of the power amplifier 3 is in general unknown, a feedback path is applied to identify the power amplifier 3 and to compute the inverse for the predistorter 1. The feedback path comprises an analog-to-digital converter (ADC) 4 for converting an output-sensed signal back into the discrete-time domain.
To compensate the out-off-band spectral components caused by the nonlinearity, the predistorter 1 can be operated at a sufficiently high sampling frequency Ωs, which is at least 2P-times higher than the transmission signal bandwidth B at the input of the power amplifier 3, where the factor P denotes the highest order of significant nonlinearity in the power amplifier 3. Therefore, the DAC 2 operates at the high sampling rate Ωs. On the other hand, the identification of a Volterra system—in
After Volterra kernel identification at low sampling rate Ωf, the identified Volterra kernels describing the nonlinear behaviour of the power amplifier 3 are used to find the correct setting of the predistorter 1. This step involves Volterra kernel upsampling.
Volterra kernel identification and Volterra kernel upsampling are illustrated in
The following notation and definitions are introduced. A continuous-time weakly nonlinear system 30, often representing nonlinear power amplifiers, will be referred to as the system V: u→y, with u, yεL2(R+), where L2(R+) denotes the space of square integrable signals and R+ is the set of positive reals. Its associated set of Laplace-domain Volterra kernels is {V(s1, . . . , sp)} with p=1, . . . , P, where the order of homogeneity of a kernel is not given explicitly but can be read off from the number of its arguments.
A discrete-time Volterra model 50 operating at the Nyquist rate Ωf of the input signal will be referred to as the low-rate model {circumflex over (V)}: û→ŷ with û, ŷεI2(Z+), where I2(Z+) denotes the space of square summable infinite sequences and Z+ is the set of positive integers. Its associated Z-transform kernels are {{circumflex over (V)}(z1, . . . , zp)} with p=1, . . . , P. The ADCs 40.1 and 40.2 provide the input for the low-rate model {circumflex over (V)} and correspond to ADC 4 in
The upsampled version of the Volterra model 50 is discrete time upsampled Volterra model 60 operating at an integer multiple of Ωf, especially at the Nyquist rate of the output signal Ωs. It will be termed the upsampled model
With a slight deviation from standard notation, but for the sake of conciseness and without loss of generality in case of band-limited input signals, it is associated with the system V its exact equivalent discrete-time system I2(Z+)→I2(Z+) at Ωs with kernels {V(z1, . . . , zp)}. The following can be use in a frequency domain analysis, i.e., sk=jΩk and zk=ejθ
Referring to kernel identification, one can find a low-rate model {circumflex over (V)} reproducing exactly the sampled output y(nTf) of the system V with Tf=2π/Ωf. This is depicted in
Upsampling of the Volterra kernels {
Note that the periodic extension implies that the kernels of the obtained upsampled model coincide with those of the system V over a limited frequency region, i.e.,
The operation of the functional elements 101, 102, 103 can be expressed by operators. The delay element 101 applies the delay operation z-δ with δ=nP, nεZ+, the nonlinear element 102 applies the nonlinear part {hacek over (V)}n of an operator {hacek over (V)} and the linear element 103 applies the inverse of the linear part {hacek over (V)}1 of the operator {hacek over (V)}. Thus, the overall function of the predistorter 100 can be expressed by the operator P with
P≡z−δ+{hacek over (V)}1−1∘{hacek over (V)}n, (1)
where “o” denotes the composition of operators. The delay of δ samples is used to compensate the latency of the cascade {hacek over (V)}1−1∘{hacek over (V)}n. It is shown later under section “Mathematical Description of Predistorter Architecture” that this specific predistorter architecture is capable of linearizing the nonlinear element 300 when selecting a proper operator {hacek over (V)} and that the kernels of this proper operator {hacek over (V)} are given by
{hacek over (V)}1(z)=−V1(z)zδ for p=1 and (2)
{hacek over (V)}p(z1, . . . , zp)=Vp(z1, . . . , zp) for p≧2. (3)
Returning to
The nonlinear element 300 is modeled by a purely linear unit 301 represented by the linear part V1 of the operator V, a nonlinear unit 302 represented by the nonlinear part Vn of the operator V and an adder 303 adding the output of the linear and nonlinear units 301, 302. The adder 303 outputs the output signal y(n).
The input of the nonlinear element 300 is down-sampled in a down-sampling stage 400.1 by the factor P and the output of the nonlinear element 300 is down-sampled in a down-sampling stage 400.2 by the factor P. It is also possible that a smaller down-sampling factor than P is used. In any case, the sampling rate at the output of the down-sampling stages 400.1, 400.2 is lower than twice the bandwidth at the output of the nonlinear element 300.
The down-sampled signals are input into an identification unit such as a low-rate kernel identification stage 700. The low-rate kernel identification stage 700 can identify a nonlinearity parameter representing the nonlinearity of the nonlinear element 300 and sample the output of the nonlinear element at an identification sampling rate that is lower than twice the bandwidth at the output of the nonlinear element. The low-rate kernel identification stage 700 comprises a low-rate nonlinear element modeling unit 701. Further, it may comprise some circuitry (adder 702, adder 703 and unit 704) adapted to minimize a cost function over the nonlinearity parameter for the purpose of identifying the nonlinear element 300.
The low-rate nonlinear element modeling unit 701 comprises a linear stage 705 operating according to the linear part {circumflex over (V)}1 of the low-rate model operator {circumflex over (V)} and a nonlinear stage 706 operating according to the nonlinear part {circumflex over (V)}n of the low-rate model operator {circumflex over (V)}. The operation characteristics of both stages 705, 706 may be changed by adjusting parameters controlling the function of the respective stages 705, 706. As there are many different possibilities to parameterize the low-rate operator parts {circumflex over (V)}1, {circumflex over (V)}n, no specific parameterization is outlined here.
During the initial identification of the nonlinear element 300, the predistorter 100 is bypassed by setting the switch 900 in position a. Identification of the nonlinear element 300 means determination of the low-rate model operator {circumflex over (V)}, i.e. to find the set of parameters for the stages 705, 706 which set the characteristic of the low-rate nonlinear element modeling unit 701 to the low-rate model operator {circumflex over (V)} reproducing exactly the sampled output y(nTf) of the nonlinear element operator V with Tf=2π/Ωf. This may be accomplished by minimizing the cost function
where y[m] is the decimated output signal of the nonlinear element 300 at the output of the down-sampling stage 400.2 and ŷ[m] is the sum of the output signals of the linear and nonlinear stages 705, 706. In other words, the input parameters of the stages 705, 706 are varied until the output of the adder 702 reaches a minimum. Although one method has been described, it will be appreciated that other methods and especially other cost functions may be applied to identify the nonlinear element 300 in the low-rate nonlinear element modeling unit 701.
In order to obtain the parameters representing the inverse {circumflex over (V)}1−1 of the linear part {circumflex over (V)}1 of the operator {circumflex over (V)}, input parameters of the inverse linear stage 704 are adjusted until the output of adder 703 reaches a minimum. Also here, other methods for obtaining the parameters representing the inverse of the linear part of the operator {circumflex over (V)} are possible.
In one embodiment, a translation unit can translate a parameter representing the nonlinearity of the nonlinear element into a predistortion control parameter. The translation unit can comprise a transformation unit that is adapted to calculate parameters which are an upsampled representation of the identified nonlinearity parameter.
The parameters calculated by the transformation unit can be calculated by applying a multi-dimensional zero padding in the time domain. In one embodiment, the translation unit includes an optimizing unit to calculate the predistortion control parameter by minimizing a second cost function over the parameters calculated by the transformation unit.
The low-rate identified operator {circumflex over (V)} is used to calculate the input parameters for the nonlinear stage 102 and the inverse linear stage 103 of the predistorter 100. Calculation of these parameters is done by obtaining the kernels of the upsampled operator
Then, the upsampled operator
{hacek over (V)}(z1, . . . , zp)=
It is to be noted that, although the upsampled operator
To see this, note that the kernels of the unsampled model (i.e. operator
To correct for the periodic extensions of {hacek over (V)}(z), one can apply a low-rate adaptation, where the initialization is taken to be the upsampled response
It is to be noted that the calculation of the predistorter parameters as described above by upsampling the nonlinear kernels of the low-rate identified nonlinear element 300 and copying these high-rate kernels into the predistorter 100 is generally not feasible for predistorter architectures.
With the applied predistortion architecture P≡z−δ+{hacek over (V)}1−1∘{hacek over (V)}n, as the operator {hacek over (V)}n with kernels {
{tilde over (P)}≡z−δ∘{1−V1−1∘[V2∘V1−1∘V2+V2∘(1−V1−1∘V2)+V3]}. (7)
In the case of upsampled kernels, every composition of operators where a nonlinear operator with k>1 is preceding one or more operators, an adaptation of all consecutive operators following Vk is necessary to yield equivalent performance to a 3rd-order inverse utilizing the correct high-rate kernels of the system V.
The block diagram shown in
Further, it is to be noted that for purely digital systems as well as for mixed signal systems, the nonlinear element 300 could be implemented by virtually any nonlinear circuitry and is not restricted to amplifiers. In addition, although described and illustrated as a hardware structure, the functionality and features of the present system can also be performed by appropriate software routines or a combination of hardware and software. Thus, the present invention should not be limited to any particular implementation.
As a practical implementation, the nonlinear element 300 may be a power amplifier for a VDSL (Very High Data Rate Digital Subscriber Line) analog front-end transmitter device that complies with the distortion ratios of the VDSL standard Band Plan 998. To be able to apply predistortion over a realistic distortion range, such circuit was miss-tuned to yield a harmonic distortion ratio of 40 dB. The operator V has been obtained by fitting a simple 5th-order Volterra model to the I/O measurement of a transistor-level circuit simulation. The standard 4-band VDSL system can utilize a bandwidth of up to 12 MHz, where each band is DMT (Discrete Multi-Tone Modulation) modulated. The frequency allocation for this simulation according to the standard Band Plan 998 comprises two downstream bands with bandwidth of each 2.59 MHz, located at center frequencies 2.225 MHz and 6.885 MHz. According to the above, the sampling rate of the digital predistortion for this system can be at least ΩS=2π×81.8 MHz. Conventionally, this directly translates to the sampling rate of the ADC 4 in the feedback path used for the parameterization of the predistorter. The proposed approach can use an ADC sampling rate of Ωf=2π×16.36 MHz resulting in a cost efficient implementation of such circuitry.
Mathematical Description of Predistorter Architecture
First, it is shown that the predistorter architecture depicted in
The composition of the predistorter 100 and the nonlinear element is given by the expression
L≡(V1+Vn)∘(z−δ+{hacek over (V)}1−1∘{hacek over (V)}n). (8)
This expression yields a linearization if ∥Vn∥<<∥V1∥ with {hacek over (V)}1=−zδ∘V1 and {hacek over (V)}n=Vn, where the operator norm corresponds to the signal norm with ∥V1∥≡sup∥u∥<1∥V1u∥. Qualitatively this becomes clear by decomposing (8) as
L≡z−δ∘V1+V1∘{hacek over (V)}1−1∘{hacek over (V)}n+Vn∘(z−δ+{hacek over (V)}1−1∘{hacek over (V)}n), (9)
where it is used that the delay operator commutes with all time-invariant operators. The above assumption on weak nonlinearity implies ∥Vn∘{hacek over (V)}1−1∘Vn∥<<∥Vn∥, such that the last term in (9) may be approximated by Vn∘(z−δ+{hacek over (V)}1−1∘{hacek over (V)}n)≈Vn∘z−δ. Thus, the overall system response is rendered as L≈z−δ∘V1, i.e. is linear.
Second, a quantitative analysis of the linearization performance of the predistorter 100 is provided by employing the Volterra series representation of the involved operators.
The set of Volterra kernels of {hacek over (V)} is denoted by {{hacek over (V)}(z1, . . . , zp)} with p=1, . . . , P. The set of kernels {L(z1, . . . , zp)} for p=0, 1, 2, . . . , P2 of L is expressed in terms of the kernels of the operators according to the decomposition in (9). For clarity of presentation, we indicate the order of homogeneity of a Volterra kernel by an additional subscript in the following derivation. Allowing for non-minimum phase systems with {hacek over (V)}1(z){hacek over (V)}1−1(z)=Z−δ, the application of the Z-domain cascade rule for discrete-time Volterra series to (9) yields
with the lower and upper index bound functions
with α(1)=1 and
respectively.
Separating terms in (10) for n=1 and n=p gives
where the first two summands in (11) equalize if the linear frequency domain kernel is {hacek over (V)}1(z)=−V1(z)zδ as already noted in (2) and the kernels for p≧2 are {hacek over (V)}p(z1, . . . , zp)+Vp(z1, . . . , zp) as already noted in (3).
In conclusion, the system with an adjusted predistorter 300 results in the following Volterra kernels
Although “perfect” equalization is possible for the 2nd-order Volterra kernel L2(z1,z2), the resulting distortion of the predistorted non-linear element output signal is much lower compared to the non-compensated non-linear element output signal if ∥Vn∥<<∥V1∥ because all the kernels Lp(z1, . . . , zp) for p≧3 in (12) incorporate the multiplicative inverse of the linear kernel V1(z).
Although the invention has been illustrated and described with respect to one or more implementations, alterations and/or modifications may be made to the illustrated examples without departing from the spirit and scope of the appended claims. In particular regard to the various functions performed by the above described components or structures (blocks, units, engines, assemblies, devices, circuits, systems, etc.), the terms (including a reference to a “means”) used to describe such components are intended to correspond, unless otherwise indicated, to any component or structure which performs the specified function of the described component (e.g., that is functionally equivalent), even though not structurally equivalent to the disclosed structure which performs the function in the herein illustrated exemplary implementations of the invention. In addition, while a particular feature of the invention may have been disclosed with respect to only one of several implementations, such feature may be combined with one or more other features of the other implementations as may be desired and advantageous for any given or particular application. Furthermore, to the extent that the terms “including”, “includes”, “having”, “has”, “with”, or variants thereof are used in either the detailed description and the claims, such terms are intended to be inclusive in a manner similar to the term “comprising”.
This application is a Continuation of co-pending application Ser. No. 11/500,107. The priority of the application Ser. No. 11/500,107 is hereby claimed, and the entirety of the application is hereby incorporated herein by reference.
Number | Name | Date | Kind |
---|---|---|---|
5732333 | Cox et al. | Mar 1998 | A |
5900778 | Stonick et al. | May 1999 | A |
6335767 | Twitchell et al. | Jan 2002 | B1 |
6342810 | Wright et al. | Jan 2002 | B1 |
6356146 | Wright et al. | Mar 2002 | B1 |
6377116 | Mattsson et al. | Apr 2002 | B1 |
6501805 | Twitchell | Dec 2002 | B1 |
6600516 | Danielsons et al. | Jul 2003 | B1 |
6775330 | Bach et al. | Aug 2004 | B2 |
6889060 | Fernando et al. | May 2005 | B2 |
7091779 | Sahlman | Aug 2006 | B2 |
7268620 | Nygren et al. | Sep 2007 | B2 |
7613251 | Shako et al. | Nov 2009 | B2 |
7688914 | Ohba et al. | Mar 2010 | B2 |
7795858 | Tufillaro et al. | Sep 2010 | B2 |
7813496 | Kubin et al. | Oct 2010 | B2 |
7885797 | Koppl et al. | Feb 2011 | B2 |
7894788 | Keehr et al. | Feb 2011 | B2 |
8294605 | Pagnanelli | Oct 2012 | B1 |
8295790 | Koren et al. | Oct 2012 | B2 |
20030223508 | Ding et al. | Dec 2003 | A1 |
20050212596 | Batruni | Sep 2005 | A1 |
20050219089 | Batruni | Oct 2005 | A1 |
20080133982 | Rawlins et al. | Jun 2008 | A1 |
20110069749 | Forrester et al. | Mar 2011 | A1 |
20110110473 | Keehr et al. | May 2011 | A1 |
Entry |
---|
Sampling Requirements for Volterra System Identification, Walter A. Frank, IEEE Signal Processing Letters, vol. 3, No. 9, Sep. 1996, pp. 266-268. |
Prosecution History from U.S. Appl. No. 11/500,107, filed Aug. 7, 2006, to include: references of Jun. 17, 2009 List of references cited by examiner; references of Jun. 17, 2009 List of References cited by applicant and considered by examiner; references of Jun. 17, 2009 List of References cited by applicant and considered by examiner; and references of Dec. 18, 2006 Information Disclosure Statement (IDS) Form (SB08). |
Number | Date | Country | |
---|---|---|---|
20090291650 A1 | Nov 2009 | US |
Number | Date | Country | |
---|---|---|---|
Parent | 11500107 | Aug 2006 | US |
Child | 12511999 | US |