GB2490749A - Linearization of an RF power amplifier using a combined FIR-IIR amplifier model - Google Patents

Abstract

Linearization of an RF power amplifier using a combined FIR-IIR amplifier model A highly non-linear RF amplifier exhibiting significant memory effects may be modelled with less use of FPGA and ASIC circuitry by using a combined FIR and IIR arrangement. The combined IIR-FIR technique requires fewer coefficients than a pure FIR technique. The RF amplifier may be a multi-stage amplifier of Doherty configuration. The modelling technique may be applied to other non-linear devices or processes.

Description

I

Modelling of Non-Linear Devices Many devices used in communications systems behave nonlinearly when op-erated at high performance or high power limits. Many of these devices can be used at low performance or low power levels to obtain linear be- havior. The behavior of such devices at linear regions can be approxi-mated with very simple mathematical formulas. These simple formulas can be used by design engineers to build complex communication systems.

However, operating in the linear region is often inefficient in power consumption as well as it increases product costs. Hence, accurate non-linear device modeling techniques are of importance as operation within the non-linear regions is an inevitable event due to above reasons. Such non-linear modeling techniques can then be used to linearize such devic-es or Power Amplifiers for instance, so that simple design techniques can be used to maximize product efficiency as well as lowering product costs. a. * a * * * S.

a novel technique is proposed in the following whereby a multi stage power amplifier of Doherty configuration can be modeled very accu- *retely at its non-linear regions. Although the technique was studied for multi stage power amplifier used in the Flexi Radio family, it can be *...used or applied in modeling any non-linear device or a non-linear pro-çess. I * * I **

-Figure 1 is a Six Carrier GSM Signal wih a signal bandwidth of 20MHz.

Although a vast array of non-linear modeling techniques have been dis- cussed iii the literature, none of them model well with the accuracy re-quirements. A non-linear modeling accuracy in excess of -GUdBc within the signal bandwidth is of interest, as shown in Figure 1.

A majority of the existing power amplifier modeling techniques utilize FIR (Finite Impulse Response) type of arrangement. This means low to higher order terms of the non-linear modeling architecture are organized similar to a FIR filter structure. These coefficients are fixed and do not exhibit time decaying characteristics.

Only a very few non-linear applications have been modeled with hR (In-finite Impulse Response) arrangement. Here the filter coefficients do exhibit time decaying characteristics. IIR modeling techniques (non-linear modeling) have been used mostly in Control Theory. Its use in modeling amplifiers in communication systems is relatively new.

*:\ese techniques were used to model a highly non-linear amplifier with :lcjrge memory. The amplifier is arranged in a Doherty configuration.

arteristics of the FIR and IIR techniques are described below: S..

S

FIR modelling: S. * * S * -Model accuracy is good, but requires a large amount of coeffi-cients to attain model accuracy in excess of -60 dBc within the signal bandwidth.

-Matrix inversion that solves the system of equations for the non-linear model may become unstable at times.

-FIR model is not suitable for modeling short or midterm transient behavior of Power Amplifiers. Hence, the primary function of the FIR based non-linear modeling technique would be to model the long term changes of the power amplifiers.

-FIR technique requires heavy use of FPGA (Field Programmable Gate Array) and ASIC (Application Specific Integrated Circuit) re-sources due to increased number of coefficients.

hR modeling: -None of the existing hR modeling techniques yield required model accuracy.

-Due to time decaying behavior of the IIR filter structure, it can model the Power Amplifier midterm transient characteristics. The-se midterm transient behavior of the power amplifiers last from micro seconds to a millisecond in duration.

a concluding remark, the existing FIR or hR techniques by itself d6es not model highly non-linear power amplifiers exhibiting large * m&mory. * * * S.. S..

Based on the deficiencies mentioned above, a combined FiR and hR model- 0ing technique is proposed whereby a highly non-linear Power Amplifier * large memory is modeled. The combined technique yields a modeling error in the order of -68 dBc to -70 dBc. In addition to the much im-proved model error, the combined technique models both the long term Power Amplifier changes as well as the mid term transient behavior.

In addition to model improvements the combined technique simplifies im-plementaticn due to relatively low use of FEGA resources compared with an FIR implementation-The combined FIR and HR technique does not re-quire a large amount of coefficients like the FIR method.

several combinations of FIR and IIR term usage can be found where addi-tional improvements in model error yielded. In the following, generic techniques are described, referring to the Figures 1, 2, 3 and 4, where FIR and IIR terms are used together to obtain -68dBc to -7OdBc model im-provements.

ía (N) Ad 2R (N) Ad 2F? E q1vI(n-k1) ÷E bx1(n-k))x1(n-k) + bxl(n-k2))xl(fl-k2_C1flS + kt-LI Ri k=O Ri=I k2=O (N) Ad flU (N) Ad ZN) (N) Al 2R4 -k3)x1(n -k3 + d2)I + c3x1(n -k)v1(n -+ cxl(n -k2)lv1(n -U -R2=I k3O R3I k=O R4=I k2O (2115) Ad 115 4x1(n -k3)v2(n -k3 + d2)I 115=1 k3O Equation 1: Generic Combination 1 �S S * * . *

S S'S..

Referring to the generic combination 1, depicted in Equation 1, a1 is a * °* Ic 1c2 k3 *linear model coefficient, b, b, and b are coefficients for non-R Ri R2 linear terms. dl and d2 are considered cross terms that range between -5 ** *5 * a * * and -s-S in general. R is the non-linear model order. S. 5 * U * * **

xl(n-k) is the input signal with a k time shift at time n. Y(n) is the power amplifier output signal at time n. xl(n-k) and Y(n) are base band signals. vl und 2 are obtained by passing xl through a single pole and a double pole HR filters respectively. Poles are optimized separately k Jc2 k3 from the non-linear adaptation process CR3 CR4, and CR5 are non-linear CoefiCient terms of the hR signals Poles are kept within the unit circle.

L2 (N) 44 21? (N) 44 2)11 E akLvl(n kl) + bx1(nk)Ixl(flk) + Ebx1(n_k2)kI(flk2_dI) ÷ kl=-LI 11=1 4=0 P1=1 42=0 (N) 44 2R2 (N) 44 2)13 (N) it! 2114 E bx1(n -k3))xl(n -k3 + d2)j + cvl(n -k)lvl(n -+ cvl(n -k2)tvl(n -k2 - 112=! 43=0 P3=1 4=0 114=! 42=0 (2115) 44 115 cv2(n -k3))v2(n -k3 + d2)$ gs=i 43=0 Equation 2: Generic Combination 2 All symbol meanings of the Generic combination 2, depicted in Equation 2, are similar to the symbols used before in Equation 1. The terms xl(n- k) lvl(n-k-d) are changed to either l(n-k) tvl(n-kd) or v2(n-k) vl(n-k-d) I. S * * * * * 0* *42 (N) 44 21? (N) 44 2)11 avl(n-k1) + bx1(n_k)Ix1(n_k)S + Ebx1(n_k2_dl)cxl(n_k2)j + 41=-LI 11=1 4=0 R1=1 k2=0 44 2112 (N) 44 2)13 (N) U 2114 bxI(n -k3 + d2)k(n -k3) + c3vx(n -k)Svx(n k) + S 5cvx(n -k2)tvx(n -k2 -dl)I R2.I 3=0 113=1 4=0 114=1 42=0 (2115) 44 115 * L.5 cvx(n -k3))'vx(n -k3 + d2)) * /L91 L=O * * ** * * S. * tcjuation 3: Generic Combination 3 Referring to the generic combination 3, depicted in Equation 3, k can take the form of either l or v2. Note that vi or v2 are obtained by passing xl through a single pole and a double pole hR filters respec-tively.

In general, it is not necessary to use polynomial base functions or to restrict the number of poles which are used for the modeling. In addi-tion to this it is not required to use one polynomial model only.

L2 1)11' I)2b (N) ;1v1(n-kl) + > *1=-LI dI=DIa d2D2u R=I P etb E2h (N) -k -e1)F2 (jv (n -k -e2* pO el-Eta e2tE2a R=l Equation 4: Generic Combination 4 In the generic combination 4, depicted in Equation 4, F( ) means func- tions from either parameter. Some of the parameters can be equal to ze-ro. *q * * ** * S.

:n9jgure 1 shows a six carrier GSM (Global System for Mobile Comrnunica- tions) signal with a signal bandwidth of 20MHz. The upper curve demon- : rate5 the multi-carrier signal at the output of the Doherty Power Am-plifier. The lower curve demonstrates the model error. The signals were :.enerated with the generic combination 1, depicted in Equation 1. * *S * C * *

S

Claims (3)

1. Claims 1. A method of modeling a non-linear device, comprising combining FIR and hR modeling.
2. 2. The method according to claim 1, wherein the device is a high-ly non-linear power amplifier.
3. 3. The method substantially as described hereinabove, with refer--ence to the accompanying drawing. S. * * * a * *4 * S * II.I *t�I "I ** * C * * I * * II