BASEBAND EQUIVALENT VOLTERRA SERIES FOR DIGITAL PREDISTORTION IN MULTI-BAND POWER AMPLIFIERS

Abstract

Methods, systems and apparatus for modelling a power amplifier and pre-distorter fed by a multi-band signal are disclosed. According to one aspect, a method includes receiving a multi-band signal and generating a discrete base band equivalent, BBE, Volterra series based on the received multi-band signal, where the series has distortion products grouped according to determined shared kernels. The shared kernels are determined based on a transformation of a real-valued continuous-time pass band Volterra series without pruning of kernels.

Description

TECHNICAL FIELD

The present invention relates to amplifiers and transmitters, and in particular to a method and system for digital pre-distortion in multi-band power amplifiers and transmitters.

BACKGROUND

Advanced modulation techniques and access technologies are enabling high speed mobile access for users. However, these techniques are increasing the complexity of the development of radio transceivers. The continued quest for flexible and dynamic networks challenges designers to develop novel radio systems capable of processing multi-band and frequency aggregated multi-standard, multi-carrier communication signals. While radio systems designers could use multiple power amplifiers (PA) 10, 12, each one dedicated to a particular radio frequency (RF) band, as shown in FIG. 1, this solution dramatically increases the deployment cost of the network and limits network flexibility. Alternatively, a more suitable solution for future communication systems is the use of a unique multi-band PA 14 to amplify combined multi-band multi-carrier and multi-standard signals, as shown in FIG. 2. This would incur lower costs for materials and more flexibility in deployment. However, this solution imposes new efficiency and linearity challenges. In fact, a single multi-band PA should provide RF performance (efficiency, gain, output power) comparable to multiple single-band PA modules. In addition, when concurrently driven with multiple signals scattered over spaced frequencies, a multi-band PA can actually aggravate the distortion problems encountered.

Previous efforts to improve the efficiency and linearity of single-band PAs, such as load (Doherty) and drain-supply (envelope tracking) modulations, have been applied to improve efficiency at the back-off region of single-band PAs. Recent studies identified sources of bandwidth limitations and devised solutions to mitigate them. Several proof-of-concept prototypes have demonstrated excellent efficiency in the back-off region over a wide range of frequencies.

On the other hand, linearization techniques, such as Digital Pre-distortion (DPD), have been applied to extend the linear region of single-band PAs. A number of DPD schemes have been developed which have demonstrated excellent linearization capability. These schemes evolved from low complexity schemes (e.g., memory-less polynomials, Hammerstein and Wiener models, memory polynomials) to more comprehensive ones (e.g., Volterra series and Artificial Neural Networks (ANN)).

In the case of the Volterra series, its application to the linearization of single-band PAs which exhibit significant memory effects was conditional on its successful pruning. This motivated researchers investigating multi--band DPD schemes to discard the Volterra series option, worrying it would lead to unmanageable and impractical solutions. Hence, most of the recent work has concentrated on efforts to generalize the previously mentioned low complexity schemes to the dual-band PA context.

A dual-band signal can be expressed as follows:

$\begin{array}{cc}\begin{array}{c}x\left(t\right)=x_1\left(t\right)+x_2\left(t\right)\\ =\mathrm{Re}\left({\tilde{x}}_1\left(t\right){}^{{\mathrm{j\omega }}_1t}+{\tilde{x}}_2\left(t\right){}^{{\mathrm{j\omega }}_2t}\right),\end{array}& \left(1\right)\end{array}$

where x(t) is the combined dual-band dual-standard signal, x₁(t) and x₂(t) are single-band multicarrier signals modulated around the angular frequencies ω₁ and ω₂, respectively, and {tilde over (x)}₁(t) {tilde over (x)}₂(t) denote the baseband envelops of x₁(t) and x₂(t), respectively.

The dual-band input signal can be represented as a broadband signal with an angular carrier frequency equal to (ω₁+ω₂)/2 as given by:

$\begin{array}{cc}\begin{array}{c}x\left(t\right)=x_1\left(t\right)+x_2\left(t\right)\\ =\mathrm{Re}\left(\left({\tilde{x}}_1\left(t\right){}^{j\frac{{\omega }_1-{\omega }_2}{2}t}+{\tilde{x}}_2\left(t\right){}^{j\frac{{\omega }_2-{\omega }_1}{2}t}\right){}^{j\frac{{\omega }_1+{\omega }_2}{2}t}\right)\\ =\mathrm{Re}\left(\tilde{x}\left(t\right)·{}^{j\frac{{\omega }_1+{\omega }_2}{2}t}\right),\end{array}& \left(2\right)\end{array}$

where {tilde over (x)}(t) is the baseband envelope of the combined signal. When the dual-band signal is amplified by a PA, the passband component of the output signal, y_pb(t), can be described as:

$\begin{array}{cc}\begin{array}{c}{y}_{\mathrm{pb}}\left(t\right)=y_1\left(t\right)+y_2\left(t\right)\\ =\mathrm{Re}\left(\left({\tilde{y}}_1\left(t\right){}^{j\frac{{\omega }_1-{\omega }_2}{2}t}+{\tilde{y}}_2\left(t\right){}^{j\frac{{\omega }_2-{\omega }_1}{2}t}\right)·{}^{j\frac{{\omega }_1+{\omega }_2}{2}t}\right)\\ =\mathrm{Re}\left({\tilde{y}}_{\mathrm{pb}}\left(t\right)·{}^{j\frac{{\omega }_1+{\omega }_2}{2}t}\right),\end{array}& \left(3\right)\end{array}$

• • where y₁(t) and y₂(t) are multicarrier output signals modulated around the angular frequencies ω₁ and ω₂ respectively, and {tilde over (y)}₁(t) {tilde over (y)}₂(t) denote the baseband envelopes of y₁(t) and y₂(t), respectively.

In the classical PA behavioral modeling approach, the PA behavior is modeled as a single-input single-output (SISO) system where the PA output {tilde over (y)}_pb(t) is a function of the PA input {tilde over (x)}(t), as given in (4):

{tilde over (y)} _pb(t)={tilde over (f)}({tilde over (x)}(t)   (4)

• • where {tilde over (f)} is the SISO describing function of the PA 16, as shown in FIG. 3. Note that the output shown in FIG. 3 is idealized. Digitization of the SISO model requires sampling both {tilde over (x)}(t) and {tilde over (y)}_pb(t) at a high frequency rate as follows:

$f_{s,\mathrm{SiSo}}\ge \left(S+5·\max \left(\frac{B_1}{2},\frac{B_2}{2}\right)\right),$

• • where B₁ and B₂ represent the bandwidths of {tilde over (x)}₁(t) and {tilde over (x)}₂(t), respectively, and S denotes the frequency spacing between the two signals

$\left(i.e.,=f_2-f_1=\frac{{\omega }_2-{\omega }_1}{2\pi }\right),$

and where f₁ and f₂ are the two bands' carrier frequencies, respectively. The factor of 5 represents the spectrum regrowth due to PA nonlinearity which is assumed equal to 5.

Alternatively, a dual-input dual-output (DIDO) approach would require a significantly lower sampling rate. In such a formulation, the PA output in each band (i.e.,{tilde over (y)}₁(t) and {tilde over (y)}₂(t), is expressed separately as a function of the two input signals' envelopes {tilde over (x)}₁(t) and {tilde over (x)}₂(t), as given by:

{tilde over (y)} ₁(t)={tilde over (f)}₁({tilde over (x)}₁(t), {tilde over (x)}₂(t))

{tilde over (y)} ₂(t0={tilde over (f)}₂({tilde over (x)}₁(t), {tilde over (x)}₂(t))   (5)

• • where {tilde over (f)}₁ and {tilde over (f)}₂ form the PA's 18 dual-band describing functions, as shown in FIG. 4. Note that the output shown in FIG. 4 is idealized. Actual output depends on the success of the pre-distortion approach employed. The construction of the two describing functions, {tilde over (g)}₁ and {tilde over (g)}₂, needed to model and/or to linearize the dual-band PA, is performed in the digital domain This requires the sampling of {tilde over (x)}₁(t), {tilde over (x)}₂(t), {tilde over (y)}₁(t) and {tilde over (y)}₂(t) at a frequency rate given by

f _s,DiDo≧(5·max(B₁, B₂))

This sampling rate is independent of the frequency separation, S, which may be very large. Hence, f_s,DiDo is significantly lower than f_s,SiSo. For example, if we assume a dual-band signal composed of a 15 MHz WCDMA signal around 2.1 GHz and a 10 MHz LTE one centered at 2.4 GHz, the theoretical sampling frequency needed for the dual-band model, f_s,DiDo, has to be at least equal to 75 MHz; significantly lower than the 675 MHz sampling frequency required for the SISO model. The ratio between the two sampling frequencies is equal to

$\frac{f_{s,\mathrm{SiSo}}}{f_{s,\mathrm{DiDo}}}=0.11.$

There have been several attempts to devise describing functions in order to implement a dual-band model as given in equation (5). Some have proposed a third order frequency selective pre-distortion technique to handle each band separately in order to model and/or linearize PAs exhibiting strong “differential” memory effects (i.e., high imbalance between the upper and lower in-band and inter-band distortion components). This technique was tested using a multi-carrier 1001 WCDMA signal and extended to address the 5^th order inter-modulation distortions of a PA driven with multi-tone signals. Although this technique was applied to multicarrier single-band signals, it can be generalized to the dual-band case provided the required sampling rate is reduced to cope with large frequency spacing.

Some have proposed an IF dual-band model implementing a Weiner-Hammerstein DPD scheme using a sub-sampling feedback path. Although the reported simulation results showed 10 dB spectrum regrowth reduction, the proposed architecture involved digital to analog conversion (DAC) and analog to digital conversion (ADC) with disproportionate sampling rates and complicated IF processing. Furthermore, starting with a 5^th order memoryless model driven with a dual-band signal, some have shown that the PA's output in each band depends on both PA input signals. This observation has been generalized to the memory polynomial model to yield a two dimension DPD (2D-DPD) model. Reported linearization results demonstrated a 12 dB improvement of the adjacent channel leakage ratio (ACLR) at the cost of a large number of coefficients. However, stability issues were reported.

Some have proposed an orthogonal representation to handle the ill-conditioning problem and numerical instability of the 2D-DPD model. Alternatively, some have proposed 2D Hammerstein and 2D Weiner models to address the large number of coefficients required by the 2D-DPD model. When applied to construct a behavioral model of a dual-band PA with a nonlinearity order equal to 5 and a memory depth equal to 5, the 2D Hammerstein and 2D Weiner models needed 40 coefficients in each band as opposed to the 2D-DPD which required 150 coefficients. However, while the 2D-DPD model has been validated as a dual-band digital pre-distorter, the application of the 2D Hammerstein and 2D Weiner models to the linearization of dual PAs is problematic and only behavioral modeling results have been reported.

Some have pointed out the implementation complexity of the 2D-DPD and have suggested a two dimensional look up table (LUT)-based representation as an alternative. This latter approach was further simplified to use single dimension LUTs. When applied to the linearization of a dual-band PA driven with dual-band signals (separated by 97 MHz), the model demonstrated an ACLR of about −45 dB, which barely passes the mask. However, the proposed DPD scheme was operated with a sampling rate equal to 153.6 MHz and consequently a large oversampling rate with a 10 MHz signal. Hardware to achieve such a large oversampling rate is costly and undesirable.

Known behavior modeling and linearization approaches have been restricted to generalizing low complexity schemes for single-band PAs. Volterra series have been avoided due to the perceived unmanageable number of coefficients and consequent complexity.

SUMMARY

Methods, systems and apparatus for modelling a power amplifier and pre-distorter fed by a multi-band signal are disclosed. According to one aspect, a method includes receiving a multi-band signal and generating a discrete base band equivalent, BBE, Volterra series based on the received multi-band signal, where the series has distortion products grouped according to determined shared kernels. The shared kernels are determined based on a transformation of a real-valued continuous-time pass band Volterra series without pruning of kernels.

According to this aspect, in some embodiments, the shared kernels are determined based on the transformation of the real-valued continuous-time pass band Volterra series by steps that include transforming the real-valued continuous time pass band Volterra series to a multi-frequency complex-valued envelope series. The multi-frequency complex-valued envelope signal is transformed to a continuous-time pass band-only series, which is then transformed to a continuous-time baseband equivalent series. The continuous-time baseband equivalent signal is discretized to produce the discrete base band equivalent Volterra series. Shared kernels are identify, each shared kernel having distortion products in common with another shared kernel. In some embodiments, transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent signal includes expressing the continuous-time pass band-only series in convolution form. The Laplace transform is applied to the convolution form to produce a Laplace domain expression, which is frequency shifted to baseband to produce a baseband equivalent expression in the Laplace domain The inverse Laplace transform is applied to the baseband equivalent expression to produce the continuous-time baseband equivalent series. In some embodiments, a number of terms in the Laplace domain expression are reduced via symmetry. In some embodiments, terms of the Laplace domain expression are grouped based on frequency intervals where distortion terms are not zero. In some embodiments, discretizing the continuous-time baseband equivalent series to produce the discrete base band equivalent Volterra series includes truncating the continuous-time baseband equivalent series to a finite non-linearity order, and expressing the truncated series as summations of non-linear distortion terms, with upper limits of the summations being memory depths assigned to each order of the non-linear distortion terms. In some embodiments, a distortion term is a group of distortion products multiplied by a shared kernel.

According to another aspect, embodiments include a digital pre-distorter (DPD) system. The system includes a Volterra series DPD modelling unit. The DPD modelling unit is configured to calculate a discrete base band equivalent, BBE, Volterra series. The series has distortion products grouped according to determined shared kernels. The shared kernels are determined based on a transformation of a real-valued continuous-time pass band Volterra series without pruning of kernels.

According to this aspect, the DPD may further comprise a power amplifier configured to produce an output in response to a multi-band input. The output of the power amplifier is provided to the Volterra series DPD modelling unit to enable the Volterra series DPD modeling unit to compute the shared kernels based on the output of the power amplifier. In some embodiments, the DPD system further comprises a transmitter observation receiver configured to sample the output of the power amplifier and provide the sampled output to the Volterra series DPD modelling unit. In some embodiments, the distortion products and their associated kernels are determined by transforming the real-valued continuous time pass band Volterra series to a discrete base band equivalent Volterra series according to a series of steps that include: transforming the real-valued continuous time pass band Volterra series to a multi-frequency complex-valued envelope series; transforming the multi-frequency complex-valued envelope signal to a continuous-time pass band-only series; transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent series; discretizing the continuous-time baseband equivalent signal to produce the discrete base band equivalent Volterra series. The shared kernels are identified such that each shared kernel has distortion products in common with another shared kernel. In some embodiments, transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent signal includes the following steps: expressing the continuous-time pass band-only series in convolution form; applying a Laplace transform to the convolution form to produce a Laplace domain expression; frequency shifting the Laplace domain expression to baseband to produce a baseband equivalent expression in the Laplace domain; and applying an inverse Laplace transform to the baseband equivalent expression to produce the continuous-time baseband equivalent series. In some embodiments, a number of terms in the Laplace domain expression are reduced via symmetry. In some embodiments, transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent signal further includes grouping terms of the Laplace domain expression based on frequency intervals where distortion terms are not zero. In some embodiments, discretizing the continuous-time baseband equivalent series to produce the discrete base band equivalent Volterra series includes: truncating the continuous-time baseband equivalent series to a finite non-linearity order; and expressing the truncated series as summations of non-linear distortion terms, with upper limits of the summations being memory depths assigned to each order of the non-linear distortion terms. In some embodiments, a distortion term is a group of distortion products multiplied by a shared kernel.

According to another aspect, embodiments include a Volterra series digital pre-distorter, DPD, modelling unit. The DPD modelling unit includes a memory module configured to store terms of a discrete base band equivalent, BBE, Volterra series. Also, a grouping module is configured to group distortion products of the series according to determined shared kernels. The DPD modelling unit also includes a shared kernel determiner configured to determine the shared kernels based on a transformation of a real-valued continuous-time pass band Volterra series without pruning of kernels. Also, a series term calculator is configured to calculate the terms of the discrete base band equivalent Volterra series, the terms being the distortion products multiplied by their respective shared kernels.

According to this aspect, in some embodiments, the BBE Volterra series terms are based on a multi-band input. In some embodiments, the multi-band input is a dual band input. In some embodiments, the shared kernel determiner is further configured to determine the shared kernels via a least squares estimate based on the multi-band input and an output of a power amplifier. In some embodiments, the kernels and distortion products are derived from the real-valued continuous-time pass band Volterra series by: transforming the real-valued continuous time pass band Volterra series to a multi-frequency complex-valued envelope series; transforming the multi-frequency complex-valued envelope signal to a continuous-time pass band-only series; transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent series; discretizing the continuous-time baseband equivalent signal to produce the discrete base band equivalent Volterra series; and identifying the shared kernels, each shared kernel having distortion products in common

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram of a known power amplification architecture using a separate power amplifier per input signal;

FIG. 2 is a block diagram of a known power amplification architecture using a single power amplifier;

FIG. 3 is an idealized signal input/output diagram for a known single input/single output system;

FIG. 4 is an idealized signal input/output diagram for a known dual input/dual output system;

FIG. 5 is a block diagram of grouping of distortion terms grouped according to whether the terms are self-distortion terms or inter-band distortion terms;

FIG. 6 is a plot of outputs of a power amplifier with no pre-distortion, with 2D DPD, and with dual band base band equivalent (BBE) Volterra series DPD for a first input signal of a first dual band input signal;

FIG. 7 is a plot of outputs of a power amplifier with no pre-distortion, with 2D DPD, and with dual band base band equivalent (BBE) Volterra series DPD for a second input signals of the first dual band input signal;

FIG. 8 is a plot of outputs of a power amplifier with no pre-distortion, with 2D DPD, and with dual band base band equivalent (BBE) Volterra series DPD for a first input signal of a second dual band input signal;

FIG. 9 is a plot of outputs of a power amplifier with no pre-distortion, with 2D DPD, and with dual band base band equivalent (BBE) Volterra series DPD for a second input signals of the second dual band input signal;

FIG. 10 is a plot of outputs of a power amplifier with no pre-distortion, with 2D DPD, and with dual band base band equivalent (BBE) Volterra series DPD for a first input signal of a third dual band input signal;

FIG. 11 is a plot of outputs of a power amplifier with no pre-distortion, with 2D DPD, and with dual band base band equivalent (BBE) Volterra series DPD for a second input signals of the third dual band input signal;

FIG. 12 is a block diagram of a digital pre-distortion power amplification system constructed in accordance with principles of the present invention;

FIG. 13 is a block diagram of a DPD modelling unit constructed in accordance with principles of the present invention;

FIG. 14 is a flowchart of an exemplary process of modelling a power amplifier fed by a multi-band input signal;

FIG. 15 is a flowchart of an exemplary process of transforming a real-valued continuous-time pass band Volterra series to a discrete base band equivalent (BBE) Volterra series;

FIG. 16 is a flowchart of an exemplary process of transforming a continuous-time pass band series to a continuous-time baseband equivalent series; and

FIG. 17 is a flowchart of an exemplary process of discretizing the continuous-time baseband equivalent series to produce the discrete base band equivalent Volterra series.

DETAILED DESCRIPTION

Before describing in detail exemplary embodiments that are in accordance with the present invention, it is noted that the embodiments reside primarily in combinations of apparatus components and processing steps related to digital pre-distortion of wideband power amplifiers fed by a multi-band signal. Accordingly, the system and method components have been represented where appropriate by conventional symbols in the drawings, showing only those specific details that are pertinent to understanding the embodiments of the present invention so as not to obscure the disclosure with details that will be readily apparent to those of ordinary skill in the art having the benefit of the description herein.

As used herein, relational terms, such as “first” and “second,” “top” and “bottom,” and the like, may be used solely to distinguish one entity or element from another entity or element without necessarily requiring or implying any physical or logical relationship or order between such entities or elements.

The Volterra series is an appropriate modeling framework for dual-band PAs which are recursive nonlinear dynamic systems with fading memory. The empirically pruned LPE Volterra series has been successfully applied to model and to linearize single-band PAs. However, application of a Volterra series to model and linearize a multi-band PA without pruning and having a manageable number of terms has not been presented. In some embodiments described herein, a BBE dual-band Volterra series formulation is derived from the original passband real-valued Volterra series to model and linearize dual-band PAs. This approach advantageously does not require pruning and the derivation set forth below is particularly attentive to addressing exponential growth in the number of coefficients experienced with the LPE approach. Thus, the present arrangement provides a method and system for modelling a power amplifier fed by a multi-band signal input. Methods described herein present a formulation of pre-distortion that transforms a continuous-time real valued Volterra series to a discrete base band equivalent Volterra series that has a reduced number of terms and is achieved without pruning. Steps for deriving the model are described below.

Step 1: Continuous-time real-valued Volterra series modeling: The Volterra series framework is initially used to describe the relationship between the real pass-band signals at the system input and output:

y(t)=Σ_p=1^NL∫_−∞^∞∫_−∞^∞h_p(τ₁, . . . , τ_p)π_j=1^px(t−τ_j)dτ_j   (6)

• • where x(t) and y(t) represent the PA input and output RF signals and NL is the nonlinearity order.

Step 2: Real-valued to complex-valued envelope signal transformation: In the case of a dual-band PA, the band-limited input signal x(t) can be expressed as:

$\begin{array}{cc}x\left(t\right)=\mathrm{Re}\left\{{\tilde{x}}_1\left(t\right){}^{{\mathrm{j\omega }}_1t}+{\tilde{x}}_2\left(t\right){}^{{\mathrm{j\omega }}_2t}\right\}=\frac{1}{2}\left({\tilde{x}}_1^{*}\left(t\right){}^{-{\mathrm{j\omega }}_1t}+{\tilde{x}}_1\left(t\right){}^{{\mathrm{j\omega }}_1t}\right)+& \left(7\right)\\ \frac{1}{2}\left({\tilde{x}}_2^{*}\left(t\right){}^{-{\mathrm{j\omega }}_2t}+{\tilde{x}}_{2\left(t\right){}^{{\mathrm{j\omega }}_2t}}\right)& \left(8\right)\end{array}$

• • where {tilde over (x)}₁(t) and {tilde over (x)}₂(t) represent the two complex baseband envelope signals that modulate the two different angular frequencies, ω₁ and ω₂. Substituting (8) into (7) yields an expression relating the output signal y(t) to {tilde over (x)}₁(t), {tilde over (x)}₂(t), ω₁ to ω₂ as follows:

y(t)=f({tilde over (x)}₁(t),{tilde over (x)}₂(t), e^±jpω1t, e^±jqω2t); (p, q) ∈ {1, . . . NL}²    (9)

• • where the describing function f is used to represent the real valued Volterra series (7). Since the output signal y(t) in (9) is a result of the application of a nonlinear function to a band-limited RF signal, it contains several spectrum components that involve multiple envelopes, {tilde over (y)}_p,q(t), which modulate the mixing products, pω₁±qω₂, of ω₁ and ω₂ as shown in (10):

$\begin{array}{cc}y\left(t\right)=\sum _{p,q=\mathrm{NL},-\mathrm{NL}}^{p+q=\mathrm{NL}}\frac{1}{2}\left({\tilde{y}}_{p,q}^{*}\left(t\right)·{}^{-j\left(p{\omega }_1+q{\omega }_2\right)t}+{\tilde{y}}_{p,q}\left(t\right)·{}^{j\left(p{\omega }_1+q{\omega }_2\right)t}\right)=\frac{1}{2}\left({\tilde{y}}_{0,0}^{*}\left(t\right)·{}^{-0jt}+{\tilde{y}}_{0,0}\left(t\right)·{}^{0jt}\right)+\frac{1}{2}\left({\tilde{y}}_{1,0}^{*}\left(t\right)·{}^{-j{\omega }_1t}\right)+\frac{1}{2}\left({\tilde{y}}_{1,0}^{*}\left(t\right)·{}^{-j{\omega }_1t}+{\tilde{y}}_{1,0}\left(t\right){}^{j{\omega }_1t}\right)+\frac{1}{2}\left({\tilde{y}}_{0,1}^{*}\left(t\right)·{}^{-j{\omega }_2t}+{\tilde{y}}_{0,1}\left(t\right){}^{j{\omega }_2t}\right)+\frac{1}{2}\left({\tilde{y}}_{2,-1}^{*}\left(t\right)·{}^{-j\left(2{\omega }_1-{\omega }_2\right)t}+{\tilde{y}}_{2,-1}\left(t\right){}^{j\left(2{\omega }_1-{\omega }_2\right)t}\right)+\frac{1}{2}\left({\tilde{y}}_{-1,2}^{*}\left(t\right)·{}^{-j\left(2{\omega }_2-{\omega }_1\right)t}+{\tilde{y}}_{-1,2}\left(t\right){}^{j\left(2{\omega }_2-{\omega }_1\right)t}\right)⋮+\frac{1}{2}\left({\tilde{y}}_{\mathrm{NL},0}^{*}\left(t\right)·{}^{\mathrm{NL}{\mathrm{j\omega }}_1t}+{\tilde{y}}_{\mathrm{NL},0}\left(t\right){}^{\mathrm{NL}{\mathrm{j\omega }}_1t}\right)+\frac{1}{2}\left({\tilde{y}}_{0,\mathrm{NL}}^{*}\left(t\right)·{}^{-\mathrm{NL}{\mathrm{j\omega }}_2t}+{\tilde{y}}_{0,\mathrm{NL}}\left(t\right){}^{\mathrm{NL}{\mathrm{j\omega }}_2t}\right)& \left(10\right)\end{array}$

• • Here, {tilde over (y)}_0,0(t) denotes the envelope at DC, {tilde over (y)}_1,0(t) and {tilde over (y)}_0,1 (t) denote the envelopes of the first order in-band signals, and {tilde over (y)}_(2, −1) (t) and {tilde over (y)}_(−1,2) (t), denote the envelopes of the third order inter-band signals. Finally, {tilde over (y)}_(0,NL) (t) and {tilde over (y)}_(NL,0) (t), represent the first and second NLth harmonics, respectively.

Step 3: Multi frequency to passband only transformation: Equating the terms on the right sides of (9) and (10) that share the same frequency range (fundamental, mixing products) yields a multi-frequency model consisting of several distinct equations that relate the output envelopes {tilde over (y)}_p,q(t) to {tilde over (x)}₁(t), {tilde over (x)}₂(t), ω₁ and ω₂. Since we are mainly interested in the relationship between the envelopes of the output and input signals around the two carriers' frequencies, only the passband components of the PA output are considered in the equation below:

$\begin{array}{cc}{y}_{\mathrm{pb}}\left(t\right)=\frac{1}{2}\left({\tilde{y}}_{{\omega }_1}^{*}\left(t\right)·{}^{-j{\omega }_1t}+{\tilde{y}}_{{\omega }_1}\left(t\right){}^{j{\omega }_1t}\right)+\frac{1}{2}\left({\tilde{y}}_{{\omega }_2}^{*}\left(t\right)·{}^{-j{\omega }_2t}+{\tilde{y}}_{{\omega }_2}\left(t\right){}^{j{\omega }_2t}\right)=\frac{1}{2}\left(y_{{\omega }_1}^{*}\left(t\right)\right)+\frac{1}{2}\left(y_{{\omega }_2}^{*}\left(t\right)+y_{{\omega }_2}\left(t\right)\right)& \left(11\right)\end{array}$

where y_ω1(t)={tilde over (y)}_ω1(t)e^jω1t and y_ω2(t)={tilde over (y)}_ω2(t)e^jω2t

Additional derivations are applied to produce the detailed expression for the first term in (11),y_ω, (0, around the first frequency Similar derivations can be used to produce the expression of the second frequency term, y_ω1(t), which can be modeled as a summation of the Volterra series nonlinear terms y_ω1,2k+1(t) of order 2k+1. The term can also be expressed as a function of the envelopes of the nonlinear terms, y_ω1,2k+1(t), denoted hereafter {tilde over (y)}_ω1,2k+1(t) and the angular frequency ω₁ as follows:

$\begin{array}{cc}{y}_{{\omega }_1}\left(t\right)=\sum _{k=0}^{\infty }y_{{\omega }_1,2k+1}\left(t\right)=\left(\sum _{k=0}^{\infty }{\tilde{y}}_{{\omega }_1,2k+1}\left(t\right)\right)·{}^{j{\omega }_1t}& \left(12\right)\end{array}$

It is worth noting that only odd powered terms are retained and even terms are discarded since they do not appear in the pass band response. Equating the terms on the right sides of expanded (9) and (12) yields a continuous BBE DIDO Volterra series that expresses y_ω,2k+1(t) as functions of {tilde over (x)}(t), and ω₁. Below is the expression for y_ω1,1(t) and y_ω1,3(t).

$\begin{array}{cc}{y}_{{\omega }_1,1}\left(t\right)={\int }_{-\infty }^{\infty }h_1\left({\tau }_1\right)·{\tilde{x}}_1\left(t-{\tau }_1\right){}^{j{\omega }_1\left(t-{\tau }_1\right)}·{\tau }_1y_{{\omega }_1,3}\left(t\right)={\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }h_3\left({\tau }_1,{\tau }_2,{\tau }_3\right)·\left({\tilde{x}}_1\left(t-{\tau }_1\right){}^{j{\omega }_1\left(t-{\tau }_1\right)}\right)\left({\tilde{x}}_1\left(t-{\tau }_2\right){}^{j{\omega }_1\left(t-{\tau }_2\right)}\right){\left(\tilde{x}\left(t-{\tau }_3\right){}^{j{\omega }_1\left(t-{\tau }_2\right)}\right)}^{*}+\left({\tilde{x}}_1\left(t-{\tau }_1\right){}^{j{\omega }_1\left(t-{\tau }_1\right)}\right){\left({\tilde{x}}_1\left(t-{\tau }_2\right){}^{j{\omega }_1\left(t-{\tau }_2\right)}\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_3\right){}^{j{\omega }_1\left(t-{\tau }_3\right)}\right)+\left({\tilde{x}}_1\left(t-{\tau }_1\right){}^{j{\omega }_1\left(t-{\tau }_1\right)}\right)\left({\tilde{x}}_1\left(t-{\tau }_2\right){}^{j{\omega }_1\left(t-{\tau }_2\right)}\right){\left({\tilde{x}}_1\left(t-{\tau }_3\right){}^{j{\omega }_2\left(t-{\tau }_3\right)}\right)}^{*}+\left({\tilde{x}}_1\left(t-{\tau }_1\right){}^{j{\omega }_1\left(t-{\tau }_1\right)}\right){\left({\tilde{x}}_2\left(t-{\tau }_2\right){}^{j{\omega }_2\left(t-{\tau }_2\right)}\right)}^{*}+\left({\tilde{x}}_1\left(t-{\tau }_1\right){}^{j{\omega }_1\left(t-{\tau }_1\right)}\right){\left({\tilde{x}}_2\left(t-{\tau }_2\right){}^{{\mathrm{j\omega }}_2\left(t-{\tau }_2\right)}\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_3\right){}^{{\mathrm{j\omega }}_2\left(t-{\tau }_3\right)}\right)+{\left({\tilde{x}}_2\left(t-{\tau }_1\right){}^{j{\omega }_2\left(t-{\tau }_1\right)}\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_2\right){}^{j{\omega }_1\left(t-{\tau }_2\right)}\right)\left({\tilde{x}}_2\left(t-{\tau }_3\right){}^{j{\omega }_2\left(t-{\tau }_3\right)}\right)+\left({\tilde{x}}_2\left(t-{\tau }_1\right){}^{{\mathrm{j\omega }}_2\left(t-{\tau }_1\right)}\right)\left({\tilde{x}}_1\left(t-{\tau }_2\right){}^{j{\omega }_1\left(t-{\tau }_2\right)}\right){\left({\tilde{x}}_2\left(t-{\tau }_3\right){}^{{\mathrm{j\omega }}_2\left(t-{\tau }_3\right)}\right)}^{*}+\left({\tilde{x}}_2\left(t-{\tau }_1\right){}^{{\mathrm{j\omega }}_2\left(t-{\tau }_1\right)}\right){\left({\tilde{x}}_2\left(t-{\tau }_2\right){}^{{\mathrm{j\omega }}_2\left(t-{\tau }_2\right)}\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_3\right){}^{{\mathrm{j\omega }}_1\left(t-{\tau }_3\right)}\right)+{\left({\tilde{x}}_2\left(t-{\tau }_1\right){}^{{\mathrm{j\omega }}_2\left(t-{\tau }_1\right)}\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_2\right){}^{{\mathrm{j\omega }}_2\left(t-{\tau }_2\right)}\right)\left({\tilde{x}}_1\left(t-{\tau }_3\right){}^{{\mathrm{j\omega }}_1\left(t-{\tau }_3\right)}\right)& \left(13\right)\\ \}·{\tau }_3{\tau }_2{\tau }_1& \left(14\right)\end{array}$

Step 4: Continuous-time Passband to baseband equivalent transformation: In order to be implementable in a digital processor with manageable complexity, a pass band model should be transformed into a baseband equivalent model. Such a model enables mimicking the RF nonlinear dynamic distortion while applying all the computations in baseband at a low sampling rate. The baseband equivalent model is obtained by frequency translating the pass band Volterra series model to baseband. For that purpose, the continuous time pass band Volterra series expressions of (13) and (14) are first rewritten in convolution form. The convolution form for) y_ω1,1(t) is given by:

y _{ω 1, 1}(t)=h₁(t)*({tilde over (x)}₁(t)e^jω1t)   (15)

As the kernel h₃ is tri-variate, h₃(τ₁, τ₂, τ₃), and the output y(t) is mono-variate, the output function is re-assigned as follows: y_ω1,3(t)=y_ω1,3(t₁, t₂, t₃)_|t1=t2=t3=ty_ω1,3(t, t, t). The convolution form is given below.

$\begin{array}{cc}{y}_{{\omega }_1,3}\left(t_1,t_2,t_3\right)=h_3\left(t_1,t_2,t_3\right)*\left\{\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right){\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)}^{*}+\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right)\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)+{\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)}^{*}\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right)\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)+\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right){\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)}^{*}+\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right){\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right)}^{*}\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)+{\left({\tilde{x}}_2\left(t_1\right){}^{{\mathrm{j\omega }}_2t_1}\right)}^{*}\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right)\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)+{\left({\tilde{x}}_2\left(t_1\right){}^{{\mathrm{j\omega }}_2t_1}\right)}^{*}\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right)\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)+{\left({\tilde{x}}_2\left(t_1\right){}^{{\mathrm{j\omega }}_2t_1}\right)}^{*}{\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right)}^{*}\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)+{\left({\tilde{x}}_2\left(t_1\right){}^{{\mathrm{j\omega }}_2t_1}\right)}^{*}\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right)\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)\right\}.& \left(16\right)\end{array}$

The application of the Laplace transform to (15) and (16) yield the following expressions:

$\begin{array}{cc}{Y}_{{\omega }_1,1}=ℒ\left(y_{{\omega }_1,1}\left(t\right)\right)=ℒ\left(h_1\left(t\right)\right) ·ℒ\left(\left({\tilde{x}}_1\left(t\right){}^{{\mathrm{j\omega }}_1t}\right)\right)=H_1\left(s\right){\tilde{X}}_1\left(s-{\mathrm{j\omega }}_1\right)Y_{{\omega }_1,3}\left(s_1,s_2,s_3\right)=ℒ\left(y_{{\omega }_1,3}\left(t_1,t_2,t_3\right)\right)=ℒ\left(h_3\left(t_1,t_2,t_3\right)\right)·\left\{ℒ\left(\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right){\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_3}\right)}^{*}\right)+ℒ\left(\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right){\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right)}^{*}\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)\right)+ℒ\left({\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)}^{*}\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right)\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)\right)+ℒ\left(\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right){\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)}^{*}\right)+ℒ\left(\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right){\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right)}^{*}\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)\right)+ℒ\left(\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right){\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right)}^{*}\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)\right)+ℒ\left({\left({\tilde{x}}_2\left(t_1\right){}^{{\mathrm{j\omega }}_1t_2}\right)}^{*}\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right)\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)\right)+ℒ\left(\left({\tilde{x}}_2\left(t_1\right){}^{{\mathrm{j\omega }}_2t_1}\right)\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right){\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)}^{*}\right)+ℒ\left(\left({\tilde{x}}_2\left(t_1\right){}^{{\mathrm{j\omega }}_2t_1}\right){\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right)}^{*}\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)\right)+ℒ\left({\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_2t_1}\right)}^{*}\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_1t_3}\right)\right)\right\}=ℒ\left(h_3\left(t_1,t_2,t_3\right)\right)·\left\{ℒ\left(\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)\right)ℒ\left(\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right)\right)ℒ\left({\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)}^{*}\right)+ℒ\left(\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)\right)ℒ\left(\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)\right)ℒ\left(\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_3}\right)\right)+ℒ\left({\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)}^{*}\right)ℒ\left(\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right)\right)ℒ\left(\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)\right)+ℒ\left(\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)\right)ℒ\left(\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right)\right)ℒ\left({\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)}^{*}\right)+ℒ\left(\left({\tilde{x}}_1\left(t_1\right){}^{{\mathrm{j\omega }}_1t_1}\right)\right)ℒ\left({\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right)}^{*}\right)ℒ\left(\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)\right)+ℒ\left({\left({\tilde{x}}_2\left(t_1\right){}^{{\mathrm{j\omega }}_2t_1}\right)}^{*}\right)ℒ\left(\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right)\right)ℒ\left(\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)\right)+ℒ\left(\left({\tilde{x}}_2\left(t_1\right){}^{{\mathrm{j\omega }}_2t_1}\right)\right)ℒ\left(\left({\tilde{x}}_1\left(t_2\right){}^{{\mathrm{j\omega }}_1t_2}\right)\right)ℒ\left({\left({\tilde{x}}_2\left(t_3\right){}^{{\mathrm{j\omega }}_2t_3}\right)}^{*}\right)+ℒ\left(\left({\tilde{x}}_2\left(t_1\right){}^{{\mathrm{j\omega }}_2t_1}\right)\right)ℒ\left({\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right)}^{*}\right)ℒ\left(\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)\right)+ℒ\left({\left({\tilde{x}}_2\left(t_1\right){}^{{\mathrm{j\omega }}_2t_1}\right)}^{*}\right)ℒ\left(\left({\tilde{x}}_2\left(t_2\right){}^{{\mathrm{j\omega }}_2t_2}\right)\right)ℒ\left(\left({\tilde{x}}_1\left(t_3\right){}^{{\mathrm{j\omega }}_1t_3}\right)\right)\right\}.=H_3\left(s_1,s_2,s_3\right)·\left\{{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right)X_1^{*}\left({\left(s_3-{\mathrm{j\omega }}_1\right)}^{*}\right)\right)+{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right){\tilde{X}}_1^{*}\left({\left(s_2-{\mathrm{j\omega }}_1\right)}^{*}\right){\tilde{X}}_1\left(s_3-{\mathrm{j\omega }}_1\right)+{\tilde{X}}_1^{*}\left({\left(s_1-{\mathrm{j\omega }}_1\right)}^{*}\right){\tilde{X}}_1\left(s_2-{\mathrm{j\omega }}_1\right){\tilde{X}}_1\left(s_3-{\mathrm{j\omega }}_1\right)+{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right){\tilde{X}}_2\left(s_2-{\mathrm{j\omega }}_2\right){\tilde{X}}_2^{*}\left({\left(s_3-{\mathrm{j\omega }}_2\right)}^{*}\right)+{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right){\tilde{X}}_2^{*}\left({\left(s_2-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_2\left(s_3-{\mathrm{j\omega }}_2\right)+{\tilde{X}}_2^{*}\left({\left(s_1-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_1\left(s_2-{\mathrm{j\omega }}_1\right){\tilde{X}}_2\left(s_3-{\mathrm{j\omega }}_2\right)+{\tilde{X}}_2\left(s_1-{\mathrm{j\omega }}_2\right){\tilde{X}}_1\left(s_2-{\mathrm{j\omega }}_1\right){\tilde{X}}_2^{*}\left({\left(s_3-{\mathrm{j\omega }}_2\right)}^{*}\right)+{\tilde{X}}_2\left(s_1-{\mathrm{j\omega }}_2\right){\tilde{X}}_2^{*}\left({\left(s_2-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_1\left(s_3-{\mathrm{j\omega }}_1\right)+{\tilde{X}}_2^{*}\left({\left(s_1-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_2\left(s_2-{\mathrm{j\omega }}_2\right){\tilde{X}}_1\left(s_3-{\mathrm{j\omega }}_1\right)\}& \left(18\right)\end{array}$

Since {tilde over (X)}₁ and {tilde over (X)}₂ are band limited signals, the third order distortion terms in (18) are non-zero only in a range of a frequency intervals. For example, {tilde over (X)}₁(s₁−jω₁){tilde over (X)}₂(s₂−jω₂){tilde over (X)}₂*((s₃−jω₂)*) is non-zero only when s₁ ∈ I₁, s₂ ∈ I₂, s₃ ∈ I_Z, where

$I_1=\left[{\omega }_1-\frac{B}{2},{\omega }_1+\frac{B}{2}\right];I_{2=}\left[{\omega }_2-\frac{B}{2},{\omega }_2+\frac{B}{2}\right]$

• • where B designates the bandwidth of the distortion term. Accordingly, one can redefine H₃(s₁, s₂, s₃) as follows:

$H_3\left(s_1,s_2,s_3\right)=\{\begin{array}{cc}{H}_{3,s}\left(s_1,s_2,s_3\right)& \mathrm{for}s_i\in I_1,i=1;2;3\\ H_{3,d1}\left(s_1,s_2,s_3\right)& \mathrm{for}s_1\in I_1,s_i\in I_2i=2;3\\ H_{3,d2}\left(s_{1,}s_2,s_3\right)& \mathrm{for}s_2\in I_1,s_i\in I_2i=1;3\\ H_{3,d3}\left(s_1,s_2,s_3\right)& \mathrm{for}s_3\in I_1,s_i\in I_2i=1;2\end{array}$

• • Hence, (18) can be rewritten as

$\begin{array}{cc}{Y}_{{\omega }_1,3}\left(s_1,s_2,s_3\right)=Y_{{\omega }_1,3,1}\left(s_1,s_2,s_3\right)+Y_{{\omega }_1,3,2}\left(s_1,s_2,s_3\right)+Y_{{\omega }_1,3,3}\left(s_1,s_2,s_3\right)+Y_{{\omega }_1,3,4}\left(s_1,s_2,s_3\right)\mathrm{where}Y_{{\omega }_1,3,1}\left(s_1,s_2,s_3\right)=H_{3,s}\left(s_1,s_2,s_3\right)·\left\{{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right){\tilde{X}}_1\left(s_2-{\mathrm{j\omega }}_1\right){\tilde{X}}_1^{*}\left({\left(s_3-{\mathrm{j\omega }}_1\right)}^{*}\right)+{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right){\tilde{X}}_1^{*}\left({\left(s_2-{\mathrm{j\omega }}_1\right)}^{*}\right){\tilde{X}}_1\left(s_3-{\mathrm{j\omega }}_1\right)+{\tilde{X}}_1^{*}\left({\left(s_1-{\mathrm{j\omega }}_1\right)}^{*}\right){\tilde{X}}_1\left(s_2-{\mathrm{j\omega }}_1\right){\tilde{X}}_1\left(s_3-{\mathrm{j\omega }}_1\right)\right\}Y_{{\omega }_1,3,2}\left(s_1,s_2,s_3\right)=H_{3,d1}\left(s_1,s_2,s_3\right)·\left\{{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right){\tilde{X}}_2\left(s_2-{\mathrm{j\omega }}_2\right){\tilde{X}}_2^{*}\left({\left(s_3-{\mathrm{j\omega }}_2\right)}^{*}\right)+{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right){\tilde{X}}_2^{*}\left({\left(s_2-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_2\left(s_3-{\mathrm{j\omega }}_2\right)\right\}& \\ Y_{{\omega }_1,3,3}\left(s_1,s_2,s_3\right)=H_{3,d2}\left(s_1,s_2,s_3\right)·\left\{\widetilde{X_2^{*}}\left({\left(s_1-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_1\left(s_2-{\mathrm{j\omega }}_1\right){\tilde{X}}_2\left(s_3-{\mathrm{j\omega }}_2\right)+{\tilde{X}}_2\left(s_1-{\mathrm{j\omega }}_2\right){\tilde{X}}_1\left(s_2-{\mathrm{j\omega }}_1\right)\widetilde{X_2^{*}}\left({\left(s_3-{\mathrm{j\omega }}_2\right)}^{*}\right)\right\}Y_{{\omega }_1,3,4}\left(s_1,s_2,s_3\right)=H_{3,d3}\left(s_1,s_2,s_3\right)·\left\{\widetilde{X_2}\left(s_1-{\mathrm{j\omega }}_2\right)\widetilde{X_2^{*}}\left({\left(s_2-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_1\left(s_3-{\mathrm{j\omega }}_1\right)+\widetilde{X_2^{*}}\left({\left(s_1-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_2\left(s_2-{\mathrm{j\omega }}_2\right)\widetilde{X_1}\left(s_3-{\mathrm{j\omega }}_1\right)\right\}& \left(19\right)\end{array}$

Exploiting the symmetry of H_3 (s_1,s_2,s_3), yields the following relation H₃ (s₁, s₂, s₃)=H₃(s₂, s₁, s₃)=H₃(s₃, s₂, s₁) ∀s₁, s₂, s₃ from which the following equalities can be deduced.

H _3,d1(s₁, s₂, s₃)=H_3,d2(s₂, s₁, s₃)=H_3,d3(s₃, s₂, s₁)

Consequently a single kernel H_3,d can be used instead of the three separate ones where only the variable order is adjusted each time as shown below:

H _3,d1(s₁, s₂, s₃)=H_3,d(s₁, s₂, s₃)

H _3,d2(s₁, s₂, s₃)=H_3,d(s₂, s₁, s₃)   (20)

H _3,d3(s₁, s₂, s₃)=H_3,d(s ₃, s₂, s₁)

Substituting (20) in (19) yields a new expression

$\begin{array}{cc}{Y}_{{\omega }_1,3,1}\left(s_1,s_2,s_3\right)=H_{3,s}\left(s_1,s_2,s_3\right)·\left\{{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right){\tilde{X}}_1\left(s_2-{\mathrm{j\omega }}_1\right){\tilde{X}}_1^{*}\left({\left(s_3-{\mathrm{j\omega }}_1\right)}^{*}\right)+{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right){\tilde{X}}_1^{*}\left({\left(s_2-{\mathrm{j\omega }}_1\right)}^{*}\right){\tilde{X}}_1\left(s_3-{\mathrm{j\omega }}_1\right)+{\tilde{X}}_1^{*}\left({\left(s_1-{\mathrm{j\omega }}_1\right)}^{*}\right){\tilde{X}}_1\left(s_2-{\mathrm{j\omega }}_1\right){\tilde{X}}_1\left(s_3-{\mathrm{j\omega }}_1\right)\right\}& \\ Y_{{\omega }_1,3,2}\left(s_1,s_2,s_3\right)=H_{3,d}\left(s_1,s_2,s_3\right)·\left\{{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right){\tilde{X}}_2\left(s_2-{\mathrm{j\omega }}_2\right){\tilde{X}}_2^{*}\left({\left(s_3-{\mathrm{j\omega }}_2\right)}^{*}\right)+{\tilde{X}}_1\left(s_1-{\mathrm{j\omega }}_1\right){\tilde{X}}_2^{*}\left({\left(s_2-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_2\left(s_3-{\mathrm{j\omega }}_2\right)\right\}Y_{{\omega }_1,3,3}\left(s_1,s_2,s_3\right)=H_{3,d}\left(s_2,s_1,s_3\right)·\left\{\widetilde{X_2^{*}}\left({\left(s_1-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_1\left(s_2-{\mathrm{j\omega }}_1\right){\tilde{X}}_2\left(s_3-{\mathrm{j\omega }}_2\right)+{\tilde{X}}_2\left(s_1-{\mathrm{j\omega }}_2\right){\tilde{X}}_1\left(s_2-{\mathrm{j\omega }}_1\right)\widetilde{X_2^{*}}\left({\left(s_3-{\mathrm{j\omega }}_2\right)}^{*}\right)\right\}Y_{{\omega }_1,3,4}\left(s_1,s_2,s_3\right)=H_{3,d}\left(s_3,s_2,s_1\right)·\left\{\widetilde{X_2}\left(s_1-{\mathrm{j\omega }}_2\right)\widetilde{X_2^{*}}\left({\left(s_2-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_1\left(s_3-{\mathrm{j\omega }}_1\right)+\widetilde{X_2^{*}}\left({\left(s_1-{\mathrm{j\omega }}_2\right)}^{*}\right){\tilde{X}}_2\left(s_2-{\mathrm{j\omega }}_2\right)\widetilde{X_1}\left(s_3-{\mathrm{j\omega }}_1\right)\right\}& \left(21\right)\end{array}$

The application of a frequency translation in Laplace domain to (17) and (21) allows for passband to baseband equivalent transformation of y_ω1,1 and y_ω1,3(t) in the time domain. The frequency translation of _jω_k is performed by replacing s, in (17) an (21) with u_i=s_i−j_{107 k}; i=1, 2, 3; k=1, 2. Hence, the application of the frequency translation to Y_ω1,1(s). Y_ω1,3(s₁, s₂, s₃). H₁(s), H_3,s(s₁, s₂, s₃), and H_3,d(s₁, s₂, s₃) yields the following baseband equivalent expressions in the Laplace domain.

$\begin{array}{cc}{\tilde{Y}}_{{\omega }_1,1}\left(u_1\right)=Y_{{\omega }_1,1}\left(u_1+{\mathrm{j\omega }}_1\right)=H_1\left(u_1+{\mathrm{j\omega }}_1\right){\tilde{X}}_1\left(u_1\right)={\tilde{H}}_1\left(u_1\right)\tilde{X}\left(u_1\right)& \left(22\right)\\ {\tilde{Y}}_{{\omega }_1,3}\left(u_1,u_2,u_3\right)={\tilde{Y}}_{{\omega }_1,3,1}\left(u_1,u_2,u_3\right)+{\tilde{Y}}_{{\omega }_1,3,2}\left(u_1,u_2,u_3\right)+{\tilde{Y}}_{{\omega }_1,3,3}\left(u_1,u_2,u_3\right)+{\tilde{Y}}_{{\omega }_1,3,4}\left(u_1,u_2,u_3\right)& \left(23\right)\\ \mathrm{where}\text{:}& \\ \begin{array}{c}{\tilde{Y}}_{{\omega }_1,3,1}\left(u_1,u_2,u_3\right)=Y_{{\omega }_1,3,1}\left(u_1+{\mathrm{j\omega }}_1,u_2+{\mathrm{j\omega }}_1,u_3+{\mathrm{j\omega }}_1\right)\\ =H_{3,s}\left(u_1+{\mathrm{j\omega }}_1,u_2+{\mathrm{j\omega }}_1,u_3+{\mathrm{j\omega }}_1\right)·\\ \{{\tilde{X}}_1\left(u_1\right){\tilde{X}}_1\left(u_2\right){\tilde{X}}_1^{*}\left({\left(u_3\right)}^{*}\right)+\\ {\tilde{X}}_1\left(u_1\right){\tilde{X}}_1^{*}\left({\left(u_2\right)}^{*}\right){\tilde{X}}_1\left(u_3\right)+\\ {\tilde{X}}_1^{*}\left({\left(u_1\right)}^{*}\right){\tilde{X}}_1\left(u_2\right){\tilde{X}}_1\left(u_3\right)\}\\ ={\tilde{H}}_{3,s}\left(u_1,u_2,u_3\right)·\\ \{{\tilde{X}}_1\left(u_1\right){\tilde{X}}_1\left(u_2\right){\tilde{X}}_1^{*}\left({\left(u_3\right)}^{*}\right)+\\ {\tilde{X}}_1\left(u_1\right){\tilde{X}}_1^{*}\left({\left(u_2\right)}^{*}\right){\tilde{X}}_1\left(u_3\right)+\\ {\tilde{X}}_1^{*}\left({\left(u_1\right)}^{*}\right){\tilde{X}}_1\left(u_2\right){\tilde{X}}_1\left(u_3\right)\}\end{array}& \\ \begin{array}{c}{\tilde{Y}}_{{\omega }_1,3,2}\left(u_1,u_2,u_3\right)=Y_{{\omega }_1,3,2}\left(u_1+{\mathrm{j\omega }}_1,u_2+{\mathrm{j\omega }}_2,u_3+{\mathrm{j\omega }}_2\right)\\ =H_{3,d}\left(u_1+{\mathrm{j\omega }}_1,u_2+{\mathrm{j\omega }}_2,u_3+{\mathrm{j\omega }}_2\right)·\\ \{{\tilde{X}}_1\left(u_1\right){\tilde{X}}_2\left(u_2\right){\tilde{X}}_2^{*}\left({\left(u_3\right)}^{*}\right)+\\ {\tilde{X}}_1\left(u_1\right){\tilde{X}}_2^{*}\left({\left(u_2\right)}^{*}\right){\tilde{X}}_2\left(u_3\right)\}\\ ={\tilde{H}}_{3,d}\left(u_1,u_2,u_3\right)·\\ \{{\tilde{X}}_1\left(u_1\right){\tilde{X}}_2\left(u_2\right){\tilde{X}}_2^{*}\left({\left(u_3\right)}^{*}\right)+\\ {\tilde{X}}_1\left(u_1\right){\tilde{X}}_2^{*}\left({\left(u_2\right)}^{*}\right){\tilde{X}}_2\left(u_3\right)\}\end{array}& \\ \begin{array}{c}{\tilde{Y}}_{{\omega }_1,3,3}\left(u_1,u_2,u_3\right)=Y_{{\omega }_1,3,3}\left(u_1+{\mathrm{j\omega }}_2,u_2+{\mathrm{j\omega }}_1,u_3+{\mathrm{j\omega }}_2\right)\\ =H_{3,d}\left(u_2+{\mathrm{j\omega }}_1,u_1+{\mathrm{j\omega }}_2,u_3+{\mathrm{j\omega }}_2\right)·\\ \{{\tilde{X}}_2^{*}\left({\left(u_1\right)}^{*}\right){\tilde{X}}_1\left(u_2\right){\tilde{X}}_2\left(u_3\right)+\\ {\tilde{X}}_2\left(u_1\right){\tilde{X}}_1\left(u_2\right){{\tilde{X}}_2^{*}\left(u_3\right)}^{*})\}\\ ={\tilde{H}}_{3,d}\left(u_2,u_1,u_3\right)·\\ \{{\tilde{X}}_2^{*}\left({\left(u_1\right)}^{*}\right){\tilde{X}}_1\left(u_2\right){\tilde{X}}_2\left(u_3\right)+\\ {\tilde{X}}_2\left(u_1\right){\tilde{X}}_1\left(u_2\right){\tilde{X}}_2^{*}\left({\left(u_3\right)}^{*}\right)\}\end{array}& \\ \begin{array}{c}{\tilde{Y}}_{{\omega }_1,3,4}\left(u_1,u_2,u_3\right)=Y_{{\omega }_1,3,4}\left(u_1+{\mathrm{j\omega }}_2,u_2+{\mathrm{j\omega }}_2,u_3+{\mathrm{j\omega }}_1\right)\\ =H_{3,d}\left(u_3+{\mathrm{j\omega }}_1,u_2+{\mathrm{j\omega }}_2,u_1+{\mathrm{j\omega }}_2\right)·\\ \{{\tilde{X}}_2\left(u_1\right){\tilde{X}}_2^{*}\left({\left(u_2\right)}^{*}\right){\tilde{X}}_1\left(u_3\right)+\\ {\tilde{X}}_2^{*}\left({\left(u_1\right)}^{*}\right){\tilde{X}}_2\left(u_2\right){\tilde{X}}_1\left(u_3\right)\}\\ ={\tilde{H}}_{3,d}\left(u_3,u_2,u_1\right)·\\ \{{\tilde{X}}_2\left(u_1\right){\tilde{X}}_2^{*}\left({\left(u_2\right)}^{*}\right){\tilde{X}}_1\left(u_3\right)+\\ {\tilde{X}}_2^{*}\left({\left(u_1\right)}^{*}\right){\tilde{X}}_2\left(u_2\right){\tilde{X}}_1\left(u_3\right)\}\end{array}& \end{array}$

The application of the inverse Laplace to (22) and (23) yields the following time domain expressions of the baseband equivalent terms:

$\begin{array}{cc}{\tilde{y}}_{{\omega }_1,1}\left(t\right)={\int }_{-\infty }^{\infty }{\tilde{h}}_1\left({\tau }_1\right)·{\tilde{x}}_1\left(t-{\tau }_1\right)·{\tau }_1{\tilde{y}}_{{\omega }_1,3,1}\left(t\right)={\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\tilde{h}}_{3,s}\left({\tau }_1,{\tau }_2,{\tau }_3\right)·\left\{\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)}^{*}+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)+{\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)\right\}·{\tau }_3{\tau }_2{\tau }_1{\tilde{y}}_{{\omega }_1,3,2}\left(t\right)={\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\tilde{h}}_{3,d}\left({\tau }_1,{\tau }_2,{\tau }_3\right)·\left\{\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_2\right)\right)}^{*}+\left({\tilde{x}}_2\left(t-{\tau }_3\right)\right)+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_2\right)\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_3\right)\right)\right\}·{\tau }_3{\tau }_2{\tau }_1{\tilde{y}}_{{\omega }_1,3,3}\left(t\right)={\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\tilde{h}}_{3,d}\left({\tau }_2,{\tau }_1,{\tau }_3\right)·\left\{+{\left({\tilde{x}}_2\left(t-{\tau }_1\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)\left({\tilde{x}}_2\left(t-{\tau }_3\right)\right)+\left({\tilde{x}}_2\left(t-{\tau }_1\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_3\right)\right)}^{*}\right\}·{\tau }_3{\tau }_2{\tau }_1{\tilde{y}}_{{\omega }_1,3,4}\left(t\right)={\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\tilde{h}}_{3,d}\left({\tau }_3,{\tau }_2,{\tau }_1\right)·\left\{+\left({\tilde{x}}_2\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_2\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)+{\left({\tilde{x}}_2\left(t-{\tau }_1\right)\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_2\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)\right\}\}·{\tau }_3{\tau }_2{\tau }_1& \left(24\right)\end{array}$

Swapping τ₂with τ₁ in {tilde over (y)}_ω1,3,3(t) and τ₃with τ₁ in {tilde over (y)}_ω1,3,4 yields the following equality

{tilde over (y)}_ω1,3,2(t)={tilde over (y)}_ω13,3(t)={tilde over (y)}_ω1,3,4(t)

Hence, the third order baseband equivalent Volterra term could be re-written as:

$\begin{array}{cc}\begin{array}{c}{\tilde{y}}_{{\omega }_1,3}\left(t\right)={\tilde{y}}_{{\omega }_1,3,1}\left(t\right)+3{\tilde{y}}_{{\omega }_1,3,2}\left(t\right)\\ ={\tilde{y}}_{{\omega }_1,3,s}\left(t\right)+{\tilde{y}}_{{\omega }_1,3,d}\left(t\right)\end{array}& \left(25\right)\end{array}$

• • where {tilde over (y)}_ω1,3,s(t)={tilde over (y)}_ω1,3,1(t) designates the PA third order single-band self-distortion term 20 and {tilde over (y)}_ω1,3,d(t)=3{tilde over (y)}_ω1,3,2(t) denotes the PA third order dual-band inter-band-distortion term 22, as shown in FIG. 5. These terms are, collectively, the baseband equivalent Volterra series which, when discretized as explained below, model the pre-distortion of a dual band power amplifier.

The same derivations were applied to construct the fifth order Volterra distortion term expression which is found to be

$\begin{array}{cc}{\tilde{y}}_{{\omega }_1,5}\left(t\right)={\tilde{y}}_{{\omega }_1,5,s}\left(t\right)+{\tilde{y}}_{{\omega }_1,5,\mathrm{d1}}\left(t\right)+{\tilde{y}}_{{\omega }_1,5\mathrm{d2}}\left(t\right)\mathrm{where}\text{}{\tilde{y}}_{{\omega }_1,5,s}\left(t\right)={\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\tilde{h}}_{5,s}\left({\tau }_1,{\tau }_2,{\tau }_3,{\tau }_4,{\tau }_5\right)·\left\{\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_4\right)\right)}^{*}{\left({\tilde{x}}_1\left(t-{\tau }_5\right)\right)}^{*}+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_4\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_5\right)\right)}^{*}+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_4\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_5\right)\right)}^{*}+{\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_4\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_5\right)\right)}^{*}+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)}^{*}{\left({\tilde{x}}_1\left(t-{\tau }_4\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_4\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_5\right)\right)+{\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_4\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_5\right)\right)+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)}^{*}{\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_4\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_5\right)\right)+{\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_4\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_5\right)\right)+{\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)}^{*}{\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_4\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_5\right)\right)\right\}·{\tau }_5{\tau }_4{\tau }_3{\tau }_2{\tau }_1{\tilde{y}}_{{\omega }_1,5,d1}\left(t\right)={\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\tilde{h}}_{5,d1}\left({\tau }_1,{\tau }_2,{\tau }_3,{\tau }_4,{\tau }_5\right)·\left\{+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)\left({\tilde{x}}_2\left(t-{\tau }_4\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_5\right)\right)}^{*}+{\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)\left({\tilde{x}}_2\left(t-{\tau }_4\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_5\right)\right)}^{*}+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right)}^{*}{\left({\tilde{x}}_2\left(t-{\tau }_4\right)\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_5\right)\right)+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_4\right)\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_5\right)\right)+{\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)}^{*}\left({\tilde{x}}_1\left(t-{\tau }_2\right)\right)\left({\tilde{x}}_1\left(t-{\tau }_3\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_4\right)\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_5\right)\right)\right\}·{\tau }_5{\tau }_4{\tau }_3{\tau }_2{\tau }_1{\tilde{y}}_{{\omega }_1,5,d2}\left(t\right)={\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }{\tilde{h}}_{5,d1}\left({\tau }_1,{\tau }_2,{\tau }_3,{\tau }_4,{\tau }_5\right)·\left\{\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)\left({\tilde{x}}_2\left(t-{\tau }_2\right)\right)\left({\tilde{x}}_2\left(t-{\tau }_3\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_4\right)\right)}^{*}{\left({\tilde{x}}_2\left(t-{\tau }_5\right)\right)}^{*}+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)\left({\tilde{x}}_2\left(t-{\tau }_2\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_3\right)\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_3\right)\right)\left({\tilde{x}}_2\left(t-{\tau }_4\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_5\right)\right)}^{*}+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_2\right)\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_3\right)\right)\left({\tilde{x}}_2\left(t-{\tau }_4\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_5\right)\right)}^{*}+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right)\left({\tilde{x}}_2\left(t-{\tau }_2\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_3\right)\right)}^{*}{\left({\tilde{x}}_2\left(t-{\tau }_4\right)\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_5\right)\right)+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_2\right)\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_3\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_4\right)\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_5\right)\right)+\left({\tilde{x}}_1\left(t-{\tau }_1\right)\right){\left({\tilde{x}}_2\left(t-{\tau }_2\right)\right)}^{*}{\left({\tilde{x}}_2\left(t-{\tau }_3\right)\right)}^{*}\left({\tilde{x}}_2\left(t-{\tau }_4\right)\right)\left({\tilde{x}}_2\left(t-{\tau }_5\right)\right)\right\}·{\tau }_5{\tau }_4{\tau }_3{\tau }_2{\tau }_1& \left(26\right)\end{array}$

Hence, the continuous-time dual-band baseband equivalent Volterra series model for each band is given by:

{tilde over (y)}_ω1(t)={tilde over (y)}_ω1,1(t)+{tilde over (y)}_ω1,3(t)+{tilde over (y)}_ω1,s(t)+ . . .   (27)

Step 5: Discrete-time baseband equivalent Volterra series model: In order to implement the dual band BBE Volterra model in a digital processor, the following signal and systems properties and approximations are used to further simplify (27).

• 1. Truncation of the Volterra model to a finite nonlinearity order NL, generally in the range of 5 to 7.
• 2. Limitation of the integral bounds (−∞, +∞) to (0, T_∞) using the signal and system causality, and the fading memory assumption (transient response time invariant Volterra series is defined as t<T_∞) └43┘. Since the impulse responses of different Volterra kernels, i.e. {tilde over (h)}_ω1,1{tilde over (h)}_ω1,3,s,{tilde over (h)}_ω1,3,d, {tilde over (h)}_ω1,5,s, . . . represent different aspects of the system, the memory spans used in the computation of the different distortion terms can be set to be different.
• 3. Using the symmetry of the terms inside the integral (Distortion components are symmetrical and Volterra kernels can be symmetrized), the number of required kernels is significantly reduced.

Digitizing the dual band BBE Volterra model yields:

$\begin{array}{cc}{\tilde{y}}_{{\omega }_1,1}\left(n\right)={\tilde{y}}_{{\omega }_1,1}\left(n\right)+{\tilde{y}}_{{\omega }_1,3}\left(n\right)+{\tilde{y}}_{{\omega }_1,5}\left(n\right)+\cdots & \left(28\right)\\ {\tilde{y}}_{{\omega }_1,1}\left(n\right)=\sum _{i_1=0}^{M_1}{\tilde{h}}_{{\omega }_1,1}\left(i_1\right){\tilde{x}}_{1,s}\left(n,i_1\right)& \\ {\tilde{y}}_{{\omega }_1,3}\left(n\right)=\sum _{i_1=0}^{M_{3,s}}\sum _{i_2=i_1}^{M_{3,s}}\sum _{i_3=i_2}^{M_{3,s}}{\tilde{h}}_{{\omega }_1,3,s}\left(i_{1,}i_2,i_3\right)·{\tilde{x}}_{3,s}\left(n,i_{1,}i_2,i_3\right)+\sum _{i_1=0}^{M_{3,d}}\sum _{i_2=i_1}^{M_{3,d}}\sum _{i_3=i_2}^{M_{3,d}}{\tilde{h}}_{{\omega }_1,3,d}\left(i_{1,}i_2,i_3\right)·{\tilde{x}}_{3,d}\left(n,i_{1,}i_2,i_3\right)& \\ {\tilde{y}}_{{\omega }_1,5}\left(n\right)=\sum _{i_1=0}^{M_{5,s}}\sum _{i_2=i_1}^{M_{5,s}}\sum _{i_3=i_2}^{M_{5,s}}\sum _{i_4=i_3}^{M_{5,s}}\sum _{i_5=i_4}^{M_{5,s}}{\tilde{h}}_{{\omega }_1,5,s}\left(i_{1,}i_2,i_3,i_4,i_5\right)·{\tilde{x}}_{5,s}\left(n,i_{1,}i_2,i_3,i_4,i_5\right)+\sum _{i_1=0}^{M_{5,d1}}\sum _{i_2=i_1}^{M_{5,d1}}\sum _{i_3=i_2}^{M_{5,d1}}\sum _{i_4=i_3}^{M_{5,d1}}\sum _{i_5=i_4}^{M_{5,d1}}{\tilde{h}}_{{\omega }_1,5,d1}\left(i_{1,}i_2,i_3,i_4,i_5\right)·{\tilde{x}}_{5,d1}\left(n,i_1,i_2,i_3,i_4,i_5\right)+\sum _{i_1=0}^{M_{5,d2}}\sum _{i_2=0}^{M_{5,d2}}\sum _{i_3=i_2}^{M_{5,d2}}\sum _{i_4=i_3}^{M_{5,d2}}\sum _{i_5=i_4}^{M_{5,d2}}{\tilde{h}}_{{\omega }_1,5,d2}\left(i_{1,}i_2,i_3,i_4,i_5\right)·{\tilde{x}}_{5,d2}\left(n,i_1,i_2,i_3,i_4,i_5\right)& \\ \mathrm{where}& \\ {\tilde{x}}_{1,s}\left(n,i_1\right)=\tilde{x}\left(n-i_1\right)& \\ {\tilde{x}}_{3,s}\left(n,i_1,i_2,i_3\right)=\widetilde{x_1}\left(n-i_1\right){\tilde{x}}_1\left(n-i_2\right){\tilde{x}}_1^{*}\left(n-i_3\right)+{\tilde{x}}_1^{*}\left(n-i_1\right){\tilde{x}}_1\left(n-i_2\right){\tilde{x}}_1\left(n-i_3\right)& \\ {\tilde{x}}_{3,d}\left(n,i_1,i_2,i_2\right)={\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_2\left(n-i_2\right){\tilde{x}}_2^{*}\left(n-i_2\right)+{\tilde{x}}_1\left(n-i_i\right){\tilde{x}}_2^{*}\left(n-i_2\right){\tilde{x}}_2\left(n-i_2\right)& \\ {\tilde{x}}_{5,s}\left(n,i_1,i_2,i_3,i_4,i_5\right)={\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_1\left(n-i_2\right){\tilde{x}}_1\left(n-i_3\right){\tilde{x}}_1^{*}\left(n-i_4\right){\tilde{x}}_1^{*}\left(n-i_s\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_1\left(n-i_2\right){\tilde{x}}_1^{*}\left(n-i_2\right){\tilde{x}}_1^{*}\left(n-i_3\right){\tilde{x}}_1\left(n-i_4\right){\tilde{x}}_1^{*}\left(n-i_5\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_1^{*}\left(n-i_2\right){\tilde{x}}_1\left(n-i_3\right){\tilde{x}}_1\left(n-i_4\right){\tilde{x}}_1^{*}\left(n-i_5\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_1\left(n-i_2\right){\tilde{x}}_1^{*}\left(n-i_3\right){\tilde{x}}_1^{*}\left(n-i_4\right){\tilde{x}}_1\left(n-i_5\right)+{\tilde{x}}_1^{*}\left(n-i_1\right){\tilde{x}}_1^{*}\left(n-i_2\right){\tilde{x}}_1^{*}\left(n-i_3\right){\tilde{x}}_1^{*}\left(n-i_4\right){\tilde{x}}_1\left(n-i_5\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_1^{*}\left(n-i_2\right){\tilde{x}}_1^{*}\left(n-i_3\right){\tilde{x}}_1\left(n-i_4\right){\tilde{x}}_1\left(n-i_5\right)+{\tilde{x}}_1^{*}\left(n-i_1\right){\tilde{x}}_1\left(n-i_2\right){\tilde{x}}_1^{*}\left(n-i_3\right){\tilde{x}}_1\left(n-i_4\right){\tilde{x}}_1\left(n-i_5\right)+{\tilde{x}}_1^{*}\left(n-i_1\right){\tilde{x}}_1^{*}\left(n-i_2\right){\tilde{x}}_1\left(n-i_3\right){\tilde{x}}_1\left(n-i_4\right){\tilde{x}}_1\left(n-i_5\right)& \\ {\tilde{x}}_{5,d1}\left(n,i_1,i_2,i_3,i_4,i_5\right)={\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_1\left(n-i_2\right){\tilde{x}}_1^{*}\left(n-i_3\right){\tilde{x}}_2\left(n-i_4\right){\tilde{x}}_1\left(n-i_5\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_1^{*}\left(n-i_2\right){\tilde{x}}_2\left(n-i_3\right){\tilde{x}}_2\left(n-i_4\right){\tilde{x}}_2^{*}\left(n-i_5\right)+{\tilde{x}}_1^{*}\left(n-i_1\right){\tilde{x}}_1\left(n-i_2\right){\tilde{x}}_1\left(n-i_3\right){\tilde{x}}_2\left(n-i_4\right){\tilde{x}}_2^{*}\left(n-i_5\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_1\left(n-i_2\right){\tilde{x}}_1^{*}\left(n-i_3\right){\tilde{x}}_2^{*}\left(n-i_4\right){\tilde{x}}_2\left(n-i_5\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_1^{*}\left(n-i_2\right){\tilde{x}}_1\left(n-i_3\right){\tilde{x}}_2^{*}\left(n-i_4\right){\tilde{x}}_2\left(n-i_5\right)+{\tilde{x}}_1^{*}\left(n-i_1\right){\tilde{x}}_1\left(n-i_2\right){\tilde{x}}_2^{*}\left(n-i_3\right){\tilde{x}}_2^{*}\left(n-i_4\right){\tilde{x}}_2\left(n-i_5\right)& \\ {\tilde{x}}_{5,d2}\left(n,i_1,i_2,i_3,i_4,i_5\right)={\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_2\left(n-i_2\right){\tilde{x}}_2\left(n-i_3\right){\tilde{x}}_2^{*}\left(n-i_4\right){\tilde{x}}_2^{*}\left(n-i_5\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_2\left(n-i_2\right){\tilde{x}}_2^{*}\left(n-i_3\right){\tilde{x}}_2\left(n-i_4\right){\tilde{x}}_2^{*}\left(n-i_5\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_2^{*}\left(n-i_2\right){\tilde{x}}_2\left(n-i_3\right){\tilde{x}}_2\left(n-i_4\right){\tilde{x}}_2^{*}\left(n-i_5\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_2\left(n-i_2\right){\tilde{x}}_2^{*}\left(n-i_3\right){\tilde{x}}_2^{*}\left(n-i_4\right){\tilde{x}}_2^{*}\left(n-i_5\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_2^{*}\left(n-i_2\right){\tilde{x}}_2\left(n-i_3\right){\tilde{x}}_2^{*}\left(n-i_4\right){\tilde{x}}_2\left(n-i_5\right)+{\tilde{x}}_1\left(n-i_1\right){\tilde{x}}_2^{*}\left(n-i_2\right){\tilde{x}}_2^{*}\left(n-i_3\right){\tilde{x}}_2\left(n-i_4\right){\tilde{x}}_2\left(n-i_5\right)& \end{array}$

In (28), M₁, M_3,s, M_3,d, M_5,s, M_s,d1 and M_3,d2 denote the memory depth of the first, third, and fifth order Volterra series distortion terms. The dual-band complex valued BBE Volterra Series in (28) includes only nonlinear distortion products of up to order 5. Expression of the dual band BBE Volterra model with higher nonlinearity can be similarly derived. In addition, only odd powered terms are retained and even terms are discarded since they do not appear in the pass band. It is worth mentioning that the distortion terms {tilde over (x)}_3,s(n, i₁, i₂, i₃) and x_5,s(n, i₁, i₂, i₃, i₄, i₅) are linear combinations of three third and ten fifth order distortion products, respectively.

According to (28) and FIG. 5, the BBE Volterra is formed by two categories of distortion terms, namely self-distortion terms 20 and inter-band band distortion terms 22. Each of these two families of terms represents a different dynamic distortion mechanism and therefore call for different values for memory depth M₁, M_3,s, M_3,d, M_5,s, M_5,d1 and M_5,d2 as shown in (28). Here, M₁, M_3,s and M_5,s represent the memory depth of the first, third and fifth order of the self-dynamic distortion terms, respectively, and M_3,d, M_5,d1 and M_5,d2 represent the memory depth of the third and fifth order inter-band dynamic distortion terms. Hence, the proposed model formulation provides the capacity to separate memory depth values in each distortion mechanism as opposed to the other approaches, such as 2D-DPD, that use a global memory depth parameter M for all of the distortion terms. This represents an additional degree of freedom for limiting the implementation complexity of the multiband BBE.

An example of the dual band BBE Volterra model of (28) is given for NL=3 and M₁=M_3,s=M_3,d=1 in (29).

{tilde over (y)}(n)={tilde over (h)}_ω1,1(0){tilde over (x)}₁(n)+{tilde over (h)}_ω1,1(1){tilde over (x)}₁(n−1)+3{tilde over (h)}_ω1,3,s(0,0,0){tilde over (x)}₁(n){tilde over (x)}₁(n){tilde over (x)}₁*(n)+{tilde over (h)}_ω1,3,2(0,0,1)(2{tilde over (x)}₁(n){tilde over (x)}₁(n−1){tilde over (x)}₁*(n)+{tilde over (x)}₁(n){tilde over (x)}₁(n){tilde over (x)}₁*(n−1))+{tilde over (h)}_ω1,3,s(0,1,1)(2{tilde over (x)}₁(n){tilde over (x)}₁(n−1){tilde over (x)}₁*(n−1)+{tilde over (x)}₁(n−1){tilde over (x)}₁(n−1){tilde over (x)}₁*(n))+3{tilde over (h)}_ω1,3,s(1,1,1){tilde over (x)}₁(n−1){tilde over (x)}₁(n−1){tilde over (x)}₁*(n−1)+

+2{tilde over (h)}_ω1,3,d(0,0,0){tilde over (x)}₁(n){tilde over (x)}₂(n){tilde over (x)}₂*(n)+{tilde over (h)}_ω1,3,d(0,0,1)({tilde over (x)}₁(n){tilde over (x)}₂(n){tilde over (x)}₂*(n−1)+{tilde over (x)}₁(n){tilde over (x)}₂*(n){tilde over (x)}₂(n−1))+2{tilde over (h)}_ω1,3,d(0,1,1){tilde over (x)}₁(n){tilde over (x)}₂(n−1){tilde over (x)}₂*(n−1)+2{tilde over (h)}_ω1,3,d(1,0,0){tilde over (x)}₁(n−1)){tilde over (x)}₂(n){tilde over (x)}₂*(n)+{tilde over (h)}_{ω1, 3,d}(1,0,1)(x{tilde over (x)}₁(n−1){tilde over (x)}₂(n){tilde over (x)}₂*(n−1)+{tilde over (x)}₁(n−1){tilde over (x)}₂*(n){tilde over (x)}₂(n−1))+2{tilde over (h)}_ω1,3,d(1,1,1){tilde over (x)}₁(n−1){tilde over (x)}₂(n−1){tilde over (x)}₂*(n−1)   (29)

A close examination of (28) reveals a number of important attributes of the dual band BBE Volterra model: Inclusion of all the possible distortion terms attributed to the static and dynamic nonlinear behavior of the PA. These involve either only the envelope of the first band signal, e.g. {tilde over (x)}₁*(n){tilde over (x)}₁²(n), and {tilde over (x)}₁(n){tilde over (x)}₁(n){tilde over (x)}₁(n−1), or result from the mixing between the two bands' envelopes, e.g. {tilde over (x)}₁(n){tilde over (x)}₂*(n){tilde over (x)}₂(n), and {tilde over (x)}₁(n){tilde over (x)}₂*(n−1){tilde over (x)}₂(n). A large number of the distortion terms included in (28) were not incorporated in the 2D-DPD model (i.e., {tilde over (x)}₂*(n){tilde over (x)}₂(n){tilde over (x)}₁(n−1) and {tilde over (x)}₂*(n−1){tilde over (x)}₂(n){tilde over (x)}₁(n−1)).

While the dual band BBE Volterra model described herein includes larger number of distortion products than other models, according to (28) these products are grouped into different sets. Each set forms a distortion term, e. g. the distortion term {tilde over (x)}_3,s(n, i₁, i₂, i₃) represents the grouping of the following distortion products {tilde over (x)}₁(n−i₁){tilde over (x)}₁(n−i₂){tilde over (x)}₁*(n−i₃), {tilde over (x)}₁(n−i₁){tilde over (x)}₁*(n−i₂){tilde over (x)}₁(n−i₃), {tilde over (x)}₁*(n−i₁){tilde over (x)}₁(n−i₂){tilde over (x)}₁(n−i₃). The distortion products that belong to a given set share the same kernel. For example, for every possible triplet (i₁, i₂, i₃) ∈ {0 . . . M}³, three 3^rd order distortion products are combined to form {tilde over (x)}_3,s(n, i₁, i₂, i₃) and consequently share one kernel {tilde over (h)}_ω1,3,s(i₂, i₃) in (28). Similarly, for every possible quintuplet (i_i, i₂, i₃, i₄, i₅) ∈ {0 . . . M}⁵, ten 5^th order distortion products are combined to form {tilde over (x)}_5,s(n, i₁, i₂, i₃, i₄, i₅) and share one kernel {tilde over (h)}_ω1,5,s(i₁, i₂, i₃, i₄, i₅) in (28). Hence, despite the fact that the models described herein involve more distortion terms, the models use comparable numbers of kernels compared to the 2D-DPD scheme.

The expression of the dual band BBE Volterra model of (28) preserves the linearity property with respect to its coefficients. Hence, the least square error (LSE) estimator can be applied to identify the kernels in (28) for a given RF PA. Equation (30) details the expression used to compute the LSE solution of (28):

A·h=Y   (30)

where A denotes the distortion products matrix, h is the kernels' vector to be estimated and Y is the vector formed by the output signal sample. Each of these variables (A, h and Y) is defined in (31) where L represents the data stream size:

$A=\left(\begin{array}{ccccccccc}{\tilde{x}}_{1,s}\left(M_1+1\right)& \cdots & {\tilde{x}}_{1,s}\left(1\right)& {\tilde{x}}_{3,s}\left(M_{3,s}+1,0,0,0\right)& \cdots & {\tilde{x}}_{3,s}\left(1,M_{3,s},M_{3,5},M_{3,s}\right)& {\tilde{x}}_{3,d}\left(M_{3,d}\_1,0,0,0,0,0\right)& \cdots & \cdots \\ ⋮& & ⋮& ⋮& & ⋮& ⋮& & ⋮\\ ⋮& \ddots & ⋮& ⋮& \ddots & ⋮& ⋮& \ddots & ⋮\\ {\tilde{x}}_{1,s}\left(L\right)& \cdots & {\tilde{x}}_{1,s}\left(L-M_1\right)& {\tilde{x}}_{3,s}\left(L,0,0,0\right)& \cdots & {\tilde{x}}_{3,s}\left(L-M_{3,s},M_{3,s},M_{3,5},M_{3,5}\right)& {\tilde{x}}_{3,d}\left(L,0,0,0,0,0\right)& \cdots & \cdots \end{array}\right)$ $h=\begin{array}{ccccc}(h_0& \cdots & h_{M_1}& h_{0,0,0}& {\cdots )}^T,\end{array}$ $Y=\begin{array}{ccc}({\tilde{y}}_{1,0}\left(M_1+1\right)& \cdots & {{\tilde{y}}_{1,0}\left(L\right))}^T\end{array}\left(31\right)$

The LSE solution to equation (30) is computed using:

{tilde over (h)}=(A^T·A)₋₁·A^T·Y   (32)

where {tilde over (h)} is the estimate of h.

To summarize the approach, a discrete base band equivalent, BBE, Volterra series is generated based on a received multi-band signal. The series has distortion products grouped according to determined shared kernels. The shared kernels are determined based on a transformation of a real-valued continuous-time pass band Volterra series without pruning of kernels. The transformation includes transforming the real-valued continuous time pass band Volterra series to a multi-frequency complex-valued envelope series. The multi-frequency complex-valued envelope signal is then transformed to a continuous-time pass band-only series. The continuous-time pass band-only signal is transformed to a continuous-time baseband equivalent series. The continuous-time baseband equivalent signal is discretized to produce the discrete base band equivalent Volterra series. Shared kernels of the discrete base band equivalent Volterra series are identified, where each shared kernel has distortion products in common with another shared kernel.

To assess the performance of the model described above, the formulation was used to model and linearize a high power dual-band RF PA. The device under test was a broadband 45 W single ended GaN PA driven with a dual-band multi-standard signal. Three test scenarios are defined:

Case 1: 20MHz (1001) WCDMA and 20 MHz LTE signals centered @ 2.1 GHz and 2.2 GHz respectively. The PAPR of the dual-band signal is equal to 10.3 dB.

Case 2: 20 MHz (1001) WCDMA and 20 MHz LTE signals centered @ 2.1 GHz and 2.4 GHz, respectively. The PAPR of the dual-band signal is equal to 10.3 dB.

Case 3: 20 MHz 4 C WCDMA and 20 MHz LTE signals centered @ 2.1 GHz and 2.8 GHz respectively. The PAPR of the dual-band signal is equal to 10.1 dB.

The proposed dual-band BBE Volterra and 2D-DPD models were each used to linearize a DUT PA. The training of the two models was conducted using samples of the PA output signals in each band sampled at 100MSPS. The nonlinearity order and memory depth of each model were individually set to achieve the best performance versus complexity trade-off in each case. For the first and third cases, the 2D-DPD and dual-band BBE Volterra parameters were set to (NL=7 and M=3) and (NL=7, M₁=2, M_3,s=M_3,d=1, M_5,s=M_5,d1=M_5,d2=0, M_7,s=0), respectively. However, in Case 2, (NL=9 and M=3) and (NL=9, M₁=2, M_3,s=M_3,d=1, M_5,s=M_5,d1 =M_5,d2=0, M₇=0, M₉=0) were found to be adequate for the 2D-DPD and the dual-band BBE Volterra, respectively.

The linearization results for the three test scenarios are shown in FIGS. 6-11 and summaries of the corresponding performances are given in Table I-III. In each of FIGS. 6-11 the results are given for no DPD, for 2D-DPD and for the linearization method described herein. In all test scenarios, the DIDO BBE Volterra model successfully linearized the PA with significantly lower complexity and slightly better performance than the 2D-DPD model. A reduction of the spectrum regrowth of about 20 dB and an ACPR of about 50 dBc were achieved by the dual-band BBE Volterra in all test scenarios.

TABLE I
Dual band standard linearization results: Case 1
	Without DPD	With Volterra DPD	With 2D-DPD
	Band 1	Band 2	Band 1	Band 2	Band 1	Band 2
	@ 2.1	@ 2.2	@ 2.1	@ 2.2	@ 2.1	@ 2.2
	GHz	GHz	GHz	GHz	GHz	GHz
Number of	0	0	20	20	84	84
coefficients
NMSE	−19	−18	−38	−37	−37	−36
(dB)
ACLR	−35	−27	−51	−48	−50	−46
(dBc)

TABLE II
Dual band standard linearization results: Case 2
	Without DPD	With Volterra DPD	With 2D-DPD
	Band 1	Band 2	Band 1	Band 2	Band 1	Band 2
	@ 2.1	@ 2.4	@ 2.1	@ 2.4	@ 2.1	@ 2.4
	GHz	GHz	GHz	GHz	GHz	GHz
Number of	0	0	25	25	135	135
coefficients
NMSE	−18	−17	−38	−36	−36	−36
(dB)
ACLR	−35	−25	−51	−47	−48	−45
(dBc)

TABLE III
Dual band standard linearization results: Case 3
	Without DPD	With Volterra DPD	With 2D-DPD
	Band 1	Band 2	Band 1	Band 2	Band 1	Band 2
	@ 2.1	@ 2.8	@ 2.1	@ 2.8	@ 2.1	@ 2.8
	GHz	GHz	GHz	GHz	GHz	GHz
Number of	0	0	20	20	84	84
coefficients
NMSE	−19	−17	−38	−35	−36	−35
(dB)
ACLR	−32	−27	−48	−47	−47	−47
(dBc)

FIG. 12 is a block diagram of a power amplification system 24 having a digital pre-distorter modelling unit 26 implementing the dual band BBE Volterra model presented herein. Note that although FIG. 12 shows only two bands, the invention is not limited to two bands but rather, can be implemented for more than two bands according to the steps described above. The power amplification system 24 includes digital pre-distorters 28a and 28b, referred to collectively as DPDs 28. The DPDs 28 receive input from the pre-distorter modelling unit 26, and pre-distort the input signals {tilde over (x)}₁ and {tilde over (x)}₂ to produce pre-distorted signals {tilde over (x)}_1p and {tilde over (x)}_2p. Each pre-distorted signal is input to a digital modulator 30 to impress the baseband signal onto a respective carrier, converted to analog by a digital (A/D) to analog converter 32, low pass filtered by a filter 34, and mixed to radio frequency (RF) by a mixer 36 to prepare the signal for amplification by an RF PA amplifier. Accordingly, the RF signals in the two paths are summed by an adder 38 and input to a power amplifier 40. A transmitter observation receiver 42 samples the output of the power amplifier 40 in each band and produces output signals y_w1 and y_w2 These output signals are used by the DPD modeling unit 26 to derive the kernel vector h according to equation (30). The DPD modelling unit 26 calculates a discrete baseband equivalent Volterra series having distortion products grouped according to determined shared kernels, where the shared kernels are based on a transformation of a real-valued continuous-time pass band Volterra series without pruning of kernels.

FIG. 13 is a more detailed view of the DPD modelling unit 26, which includes a memory module 44 in communication with a processor 46. The DPD modelling unit 26 receives the input signals {tilde over (x)}₁ and {tilde over (x)}₂ and output signals {tilde over (y)}_ω1 and {tilde over (ω)}_ω2 from the transmitter observation receiver 42 and derives the modelling vector h according to equation (32). The modelling vector h. is input to the DPDs 28 to pre-distort the input signals {tilde over (x)}₁ and {tilde over (x)}₂ to produce pre-distorted signals {tilde over (x)}_1p and {tilde over (x)}_2p. The processor 46 includes a grouping module 50, a shared kernel determiner 52 and a series term computer 54. The grouping module is configured to group distortion products of the series according to determined shared kernels. The shared kernel determiner is configured to determine the shared kernels based on a transformation of a real-valued continuous-time pass band Volterra series without pruning of kernels. The series term calculator is configured to calculate the terms of the discrete base band equivalent Volterra series, the terms being the distortion products multiplied by their respective shared kernels. The memory module 44 is configured to store terms of the discrete base band equivalent (BBE) Volterra series 48, generated by the processor 46.

FIG. 14 is a flowchart of a process for modelling a power amplifier 40 fed by a multi-band input signal. A multi-band signal is received by a digital pre-distorter 28 (block S100). A discrete BBE Volterra series is generated by the DPD modelling unit 26 based on the received multi-band input signal (block S102). The series has distortion products that are grouped by the grouping module 50 according to determined shared kernels determined by shared kernel determiner 52. The shared kernels are determined based on a transformation of a real-valued continuous-time pass band Volterra series without pruning of kernels.

FIG. 15 is a flowchart of a process for transforming a real-valued continuous time pass band Volterra series to a discrete base band equivalent Volterra series in which shared kernels are identified as set out in block S102 of FIG. 14. The real-valued continuous time pass band Volterra series is transformed to a multi-frequency complex-valued envelope series (block S104). The multi-frequency complex-valued envelope signal is transformed to a continuous-time pass band-only series (block S106). The continuous-time pass band-only signal is transformed to a continuous-time baseband equivalent series (block S108). The continuous-time baseband equivalent signal is discretized to produce the discrete base band equivalent Volterra series (block S110). Shared kernels of the discrete base band equivalent Volterra series are identified, where a shared kernel has distortion products in common with another shared kernel (block S112).

FIG. 16 is a flowchart of a process of transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent signal as shown in block S108 of FIG. 15. The continuous-time pass band-only series is expressed in convolution form (block S114). Then, the Laplace transform is applied to the convolution form to produce a Laplace domain expression (block S116). A number of terms in the Laplace domain expression may be reduced based on symmetry (block S118). The Laplace domain expression is frequency-shifted to baseband to produce a baseband equivalent expression in the Laplace domain (block S120). An inverse Laplace transform is applied to the baseband equivalent expression to produce the continuous-time baseband equivalent series (block S122).

FIG. 17 is a flowchart of a process of discretizing the continuous-time baseband equivalent series to produce the discrete base band equivalent Volterra series, as shown in block S110 of FIG. 15. The process includes truncating the continuous-time baseband equivalent series to a finite non-linearity order (block S124). The process also includes expressing the truncated series as summations of non-linear distortion terms, with upper limits of the summations being memory depths assigned to each order of the non-linear distortion terms (block S126).

Thus, a dual band BBE Volterra series-based behavioral model has been described herein to mimic and linearize the dynamic nonlinear behavior of a concurrently driven dual-band amplifier. Starting with a real-valued, continuous-time, pass band Volterra series and using a number of signal and system transformations, a low complexity complex-valued, and discrete BBE Volterra formulation was derived. While the formulation presented herein includes all possible distortion terms, it involved fewer kernels than its 2D-DPD counterpart. The model is successfully applied to digitally predistort and linearize a dual-band 45 Watt class AB GaN PA driven with different dual-band dual-standard test signals. For each band, the model used less than 25 coefficients to reduce the ACLR by up to 25 dB.

It will be appreciated by persons skilled in the art that the present invention is not limited to what has been particularly shown and described herein above. In addition, unless mention was made above to the contrary, it should be noted that all of the accompanying drawings are not to scale. A variety of modifications and variations are possible in light of the above teachings without departing from the scope and spirit of the invention, which is limited only by the following claims

Claims

1. A method of modelling a power amplifier fed by a multi-band signal input, the method comprising: receiving a multi-band signal;generating a discrete base band equivalent, BBE, Volterra series based on the received multi-band signal, the series having distortion products grouped according to determined shared kernels; andthe shared kernels being determined based on a transformation of a real-valued continuous-time pass band Volterra series without pruning of kernels.

2. The method of claim 1, wherein the shared kernels are determined based on the transformation of the real-valued continuous-time pass band Volterra series by: transforming the real-valued continuous time pass band Volterra series to a multi-frequency complex-valued envelope series;transforming the multi-frequency complex-valued envelope signal to a continuous-time pass band-only series;transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent series;discretizing the continuous-time baseband equivalent signal to produce the discrete base band equivalent Volterra series; andidentifying the shared kernels, each shared kernel having distortion products in common with another shared kernel.

3. The method of claim 2, wherein transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent signal includes: expressing the continuous-time pass band-only series in convolution form;applying a Laplace transform to the convolution form to produce a Laplace domain expression;frequency shifting the Laplace domain expression to baseband to produce a baseband equivalent expression in the Laplace domain; andapplying an inverse Laplace transform to the baseband equivalent expression to produce the continuous-time baseband equivalent series.

4. The method of claim 3, wherein a number of terms in the Laplace domain expression are reduced via symmetry.

5. The method of claim 3, further comprising grouping terms of the Laplace domain expression based on frequency intervals where distortion terms are not zero.

6. The method of claim 2, wherein discretizing the continuous-time baseband equivalent series to produce the discrete base band equivalent Volterra series includes: truncating the continuous-time baseband equivalent series to a finite non-linearity order; andexpressing the truncated series as summations of non-linear distortion terms, with upper limits of the summations being memory depths assigned to each order of the non-linear distortion terms.

7. The method of claim 6, wherein a distortion term is a group of distortion products multiplied by a shared kernel.

8. A digital pre-distorter (DPD) system, comprising: a Volterra series DPD modelling unit, the DPD modelling unit configured to: calculate a discrete base band equivalent, BBE, Volterra series, the series having distortion products grouped according to determined shared kernels; andthe shared kernels being determined based on a transformation of a real-valued continuous-time pass band Volterra series without pruning of kernels.

9. The DPD system of claim 8, further comprising: a power amplifier, the power amplifier configured to produce an output in response to a multi-band input, the output of the power amplifier provided to the Volterra series DPD modelling unit to enable the Volterra series DPD modeling unit to compute the shared kernels based on the output of the power amplifier.

10. The DPD system of claim 8, further comprising a transmitter observation receiver configured to sample the output of the power amplifier and provide the sampled output to the Volterra series DPD modelling unit.

11. The DPD system of claim 8, wherein the distortion products and their associated kernels are determined by: transforming the real-valued continuous time pass band Volterra series to a multi-frequency complex-valued envelope series;transforming the multi-frequency complex-valued envelope signal to a continuous-time pass band-only series;transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent series;discretizing the continuous-time baseband equivalent signal to produce the discrete base band equivalent Volterra series; andidentifying the shared kernels, each shared kernel having distortion products in common with another shared kernel.

12. The DPD system of claim 11, wherein transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent signal includes: expressing the continuous-time pass band-only series in convolution form;applying a Laplace transform to the convolution form to produce a Laplace domain expression;frequency shifting the Laplace domain expression to baseband to produce a baseband equivalent expression in the Laplace domain; andapplying an inverse Laplace transform to the baseband equivalent expression to produce the continuous-time baseband equivalent series.

13. The DPD system of claim 12, wherein a number of terms in the Laplace domain expression are reduced via symmetry.

14. The DPD system of claim 12, wherein transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent signal further comprises grouping terms of the Laplace domain expression based on frequency intervals where distortion terms are not zero.

15. The DPD system of claim 11, wherein discretizing the continuous-time baseband equivalent series to produce the discrete base band equivalent Volterra series includes: truncating the continuous-time baseband equivalent series to a finite non-linearity order; andexpressing the truncated series as summations of non-linear distortion terms, with upper limits of the summations being memory depths assigned to each order of the non-linear distortion terms.

16. The DPD system of claim 15, wherein a distortion term is a group of distortion products multiplied by a shared kernel.

17. A Volterra series digital pre-distorter, DPD, modelling unit, comprising: a memory module, the memory module configured to store terms of a discrete base band equivalent, BBE, Volterra series;a grouping module, the grouping module configured to group distortion products of the series according to determined shared kernels;a shared kernel determiner, the shared kernel determiner configured to determine the shared kernels based on a transformation of a real-valued continuous-time pass band Volterra series without pruning of kernels; anda series term calculator, the series term calculator configured to calculate the terms of the discrete base band equivalent Volterra series, the terms being the distortion products multiplied by their respective shared kernels.

18. The Volterra series DPD modelling unit of claim 17, wherein the BBE Volterra series terms are based on a multi-band input.

19. The Volterra series DPD modelling unit of claim 18, wherein the multi-band input is a dual band input.

20. The Volterra series DPD modelling unit of claim 17, wherein the shared kernel determiner is further configured to determine the shared kernels via a least squares estimate based on the multi-band input and an output of a power amplifier.

21. The Volterra series DPD modelling unit of claim 17, wherein the kernels and distortion products are derived from the real-valued continuous-time pass band Volterra series by: transforming the real-valued continuous time pass band Volterra series to a multi-frequency complex-valued envelope series;transforming the multi-frequency complex-valued envelope signal to a continuous-time pass band-only series;transforming the continuous-time pass band-only signal to a continuous-time baseband equivalent series;discretizing the continuous-time baseband equivalent signal to produce the discrete base band equivalent Volterra series; andidentifying the shared kernels, each shared kernel having distortion products in common.