GB2394374A - An IQ feedback predistortion loop comprising a power amplifier (PA) and a PA model - Google Patents

Abstract

There is provided an apparatus and a method for allowing the implementation of an IQ feedback predistortion loop through the use of a subband receiver with an adjustable centre frequency. The method provides: a predistortion device that predistorts a waveform when applied thereto; a first chain, which includes a hopping receiver and a real power amplifier (PA); and a second chain, which includes a modeller for modelling at least one component of the first chain, e.g. a PA model modelling the PA. The waveform received from the hopping receiver is compared with the same input waveform after the PA model has been applied: the difference between waveforms (the EVM value) being a measure of how accurately the PA model represents the real power amplifier. The loop is completed by examining the difference between EVM values and feeding back the difference value to an LUT updater in order to adjust the values held in the predistorting LUT.

Description

IMPROVEMENTS IN OR RELATING TO RF TRANSMITTERS

The present invention relates to improvements in or relating to radio frequency (RF) transmitters. In particular, the invention relates to improv 5 ing the efficiency of power amplifiers in the base station apparatus of radio telecommunications systems.

In radio telecommunications systems, high power base stations are used to establish connections to a plurality of mobile units (handsets). The new 2.5G and 3rd generation (3G) telecommunications systems, such as lo GPRS and UMTS, demand certain features in the base stations. Notably 2.5G and 3G systems require the base stations to use high power amplifi ers. Power amplifiers (PA) are used in both base stations and mobile handsets to amplify input signals. Input signals amplified by PAs may be 15 multicarrier signals, but input signals can be as simple as a two-tone signal having tones at two distinct frequencies, f, and f2.

The PAs used in base stations must be robust at high power levels.

Substantially linear transfer characteristics are considered important to the provision of robust high power amplifiers (HPA).

to An ideal linear amplifier would give an amplified version of an input signal, which at every point in its operating range has been amplified by a constant factor.

Figure 1 shows a schematic graph comparing the characteristic of an ideal linear system with that of an analogous non-linear system: the system 25 being a power amplifier arrangement for example. The abscissa represents nominally the magnitude of the input to the power amplifier at a particular instant and resultant magnitude of the output is shown on the ordinate. The

ideal linear system is characterized by the complete line A. In practice, the real characteristics of an amplifier are likely to be non-linear as represented by the broken or dashed line B. As can be seen saturation occurs at high magnitudes of input such that increasing input does not result in a corre 5 spending increase in the output signal.

One result of the use of non-ideal power amplifiers can be the ap pearance of intolerable levels of side band distortion in the output signals.

Distortion in PAs may be both amplitude distortion and phase dis tortion. An amplifier may cause an amplitude modulation to phase modula o lion (AM/PM) transfer characteristic, whereby phase variations in the out put amplified signal are dependent upon amplitude variations in the input signal. Distortion may also be purely or partly AM/AM in nature.

Distortion is a consequence of the fundamental non-linearity of the components, and may be influenced by other physical factors, including Is changes in the operational characteristics of the component (i.e. operating frequency), temperature variations, power supply fluctuations and load mix-matches. In the absence of PAs with perfect linear transfer characteristics, some non-linear distortion effects are to be expected. Distortion effects can to appear as spurious signals having frequencies which are generally in simple arithmetic selection as input frequencies; for example harmonic distortion and intermodulation distortions (IMD).

Intermodulation and harmonic distortions are important classes of ef fects generally termed "mixing products".

25 For the purposes of the following discussion, mixing products can be characterized in terms of their orders. The "order" of a mixing product, f, is given by the sum: 0(f) = |m| + |n| + + 1Z

where f= mf + nf2 + + zfi Thus the third harmonic of f,, Off, is of order three; so too is the IMD product (off - f2). A short hand for 3rd order intermodulation distortion product, IM3, will be adopted hereafter.

5 In the high power amplifier generally used in broadband RF commu nications systems, the presence of IMD is highly unwelcome. Amplifica tion of the multicarrier signals of 2.5G and 3G systems leads to a plethora of IMD products as each channel can potentially mix with every other channel. lo In response to non-linear transfer characteristics in PAs, it is known to seek to compensate for non-linearities. The apparatus for compensating for non-linear transfer characteristics is variously termed 'predistorter', lineariser' and 'equaliser'. The difference between terms is one of empha sis and order of application: a predistorter being an apparatus for applying a 15 predistortion that seeks to complement any distortion introduced by the component before the input signal is fed to the component. The term 'line ariser' emphasises the need to bring the combined lineariser and PA ar rangement as close as possible to an ideal linear PA, whereas an 'equal iser' has the goal of flattening the distortion across an operating spectrum.

to All compensating apparatus share the feature that they seek to apply a compensating function to counter the distorting effects of PAs. The com pensating function may be viewed as an approximation to the inverse or complimentary function to the non-linear transfer function associated with the PA.

25 The inverse function can be modelled in a variety of ways. In one example an arrangement of diodes is provided, the arrangement approxi mating the inverse of the distortion effects in the PA. In further examples

software is used to emulate the effect of hardware predistortion devices in real time.

Both the non-linear transfer function and the complementary predis tortion function may be approximated by the superposition of curves.

5 It is further remarked that compensating apparatus is generally im plemented within either a feed-forward (open) or a feed-back (closed) cir cuit arrangement. Much of the following discussion relates to feed-back arrangements. Known linearization techniques applied to power amplifiers essen o tially consist of three parts: a non-linear device i.e. an RF amplifier; an out put signal measurement subsystem; and a linearization subsystem, for ex ample, adaptive correction linearization or predistortion. In order to appre ciate the invention prior art linearization methods are briefly described.

One solution to the linearization is by means of so called feed for s ward linearization which is illustrated schematically in Figure 2. Such an implementation includes a cancellation loop, feed forward loop and a measurement subsystem; the latter being essentially a selective pilot re ceiver measuring the power spectral density (PSD) of the residual tone at the output. The key aspect of this method is that the PSD is measured and to hence large dynamic range is achieved. The measurement subsystem is based on pilot signal and not on a graphical method. This solution achieves a very high linearity and excellent signal to noise ratios in excess of 70 dB.

The disadvantage of the pilot is that there is poor correlation between the minima and linearity and ACPR (Adjacent Channel Power Ratio). This can 25 be improved by a spread pilot scheme but this is complex. Another disad vantage of feed forward technique is the poor efficiency due to coupler de lay line losses and the need for an error amplifier.

Another method of providing linearization is that of adaptive ana logue predistortion. Figure 3 shows a schematic representation of an adap tive analogue predistortion system. This shows a power amplifier, the out put of which is additionally input into a complex demodulator. Thereafter s coefficients of a polynomial, which represent the inverse characteristic of the power amplifier, are computed from the output of the demodulator.

The polynomial thereby generated is applied to incoming signals before they are input into the power amplifier. In this way, the input signal is modified such that attenuated parts of the output signal are amplified and lo phase adjusted to compensate for the non-linear effects; the input signal to the power amplifier is predistorted. There are however, three problems as sociated with this type of predistortion. Firstly, this system necessitates the use of a complex down-converter. The complex down sampler has less dynamic range than that of the PSD measurement receiver because the full 15 signal bandwidth has to be digitised instantaneously. This is an issue par ticularly for W-CDMA (Wideband Code Division Multiple Access) appli cation which has higher bandwidth than an EDGE carrier. Secondly, be cause the polynomial generator comprises analogue circuitry, it is inher ently imperfect especially for high order terms due to effects such as delay to mismatch and non-linearity. These imperfections therefore limit the band width and the quality of curve fit, hence degrading the linearity. Thirdly it has been found that polynomial generation does not always achieve good fit and in some cases may become unstable and produce ripple effects through the various sections of the characteristics.

25 Digital predistortion is also known. Prior art digital predistortion

systems again require a complex down-converter in the measurement sub system. This restricts the dynamic range, especially for broadband sys tems. Figure 4 shows a schematic block diagram of a known digital pre

distortion system. In such prior art systems spline and rational polynomials

are used for linearising the amplifier. The nature of predistortion demands a method of applying inverse transfer function characteristics to the ampli fer source data. The predistortion has been applied in this method by 5 means of determining the coefficients of a polynomial representing the in-

verse characteristic, and points on the curve are stored in Look-Up Tables (LUT) generator. However there are problems with achieving stable poly-

nomials that correctly characterize a typical amplifier inverse transfer char-

acteristic. lo There is not always a deterministic relationship between PSD and the characteristics of the non-linear device. Therefore the optimum predistor tion, which is the inverse of the non-linear characteristics, cannot always be readily calculated.

From the discussion of known linearization techniques above, it is 15 worth underlining that, amongst other considerations, an IQ feed-back pre-

distortion loop needs to receive the full digital transmit band in order to operate satisfactorily. The need for access to the full bandwidth conflicts with practicalities. When compensating for non-linear behaviour in a PA, a practical compensating apparatus will generally have to work under con o straints imposed by conversion between analogue and digital domains, these constraints impinge on the available bandwidth. Furthermore consid-

eration must be given to the consequences of anti-aliasing upon effective receive bandwidth.

Conversion between analogue and digital domains introduces rate 25 mixmatches: in general, ADC sample rates tend to lag behind DAC sample rates. Anti-aliasing requirements result in an additional loss of effective receive bandwidth. In the light of the foregoing, it is not possible to use conventional techniques to receive the full digital transmit band simultane

ously with enough dynamic range to detect and suppress intermodulation components. It is therefore an object of the invention to obviate or at least mitigate the aforementioned problems. In particular it is an object of the present 5 invention to allow predistortion with IQ feedback.

In accordance with one aspect of the present invention, there is pro-

vided a compensating apparatus for compensating for non-linear effects caused by one or more electronic components in a first signal path, the compensating apparatus comprising: lo a predistortion device that operates upon incoming signals, generates a predistorted signal and feeds said predistorted signal into said one or more electronic components; a monitoring receiver that generates a first output signal, the receiver being provided in the first signal path and coupled to said one or more 15 electronic components; a processing device that provides a second signal path, the process ing device receiving said predistorted signal, being arranged to generate a second output signal that simulates the non- linear effects of components in the first signal path; and to an error measurement device, which compares the first output signal and the second output signal and outputs an error value in dependence thereupon; wherein, for each of a plurality of iterations, the error value is fed into the processing device, the processing device calculates an updated 25 simulation in accordance with the error value and the predistortion device is fed with updated simulation data by the processing device; and wherein the monitoring receiver is tuneable to a predetermined cen tre frequency and receives signals over a restricted sub-band, the receiver

being operated, during each iteration, at a plurality of centre frequencies, thereby providing sufficient coverage of the full signal bandwidth present in the incoming signals.

Advantageously the compensating apparatus includes a receiver 5 modelling unit, for simulating non-linear effects occurring within the .. montonng receiver.

In accordance with another aspect of the present invention, there is provided a compensating apparatus for compensating for non-linear effects caused by one or more electronic components, the compensating apparatus . 10 compnsmg: a sub-band receiver having an adjustable centre frequency; and an IQ predistortion feedback circuit; wherein the feedback circuit iteratively samples the output of said one or more electronic components, calculates an appropriate predistortion Is factor, and applies said predistortion factor to incoming signals before said incoming signals are fed to said one or more electronic components; and wherein the sub-band receiver is operated, during a single iterative loop of the feedback circuit, at a plurality of centre frequencies in order to provide sufficient coverage of the full signal bandwidth present in the in o coming signals.

The present invention therefore seeks to allow predistortion with IQ feedback through the use of a sub-band receiver with a programmable cen tre frequency.

The invention may take the form of an RF transmitter arrangement 25 comprising: a compensating apparatus as described earlier and one or more electronic components, said one or more electronic components including a power amplifier.

In accordance with a further aspect of the present invention, there is provided a method for compensating for non-linear effects caused by an electronic component, the electronic component receiving a predistorted incoming signal and generating an output signal, the method comprising: 5 providing a sampled signal, by sampling the output signal; inputting said sampled signal into a receiver, the receiver being ad justable to a predetermined sub-band centred upon a centre frequency; selecting a centre frequency at which the receiver is to be operated; mixing the sampled signal with a mixing signal at a predetermined lo mixing frequency; filtering the output of the mixing step to allow signals in the prede termined sub-band to pass; converting the filtered, sampled signal to generate a digital sample signal; is digitally filtering the digital sample signal to generate a digitally fil tered sample signal that occupies a restricted range of frequencies; generating a model of the transfer characteristic of the electronic component; processing the input signal in accordance with the model, thereby to generating a model signal that simulates the non-linear effects caused by the component; digitally filtering the model signal to generate a digitally filtered model signal, which represents the transfer characteristic over substantially the same range of frequencies as occur in the digitally filtered sample sig 25 nal; iteratively, calculating the error vector magnitude between the digi tally filtered sample signal and the filtered model signal and updating the

model of the component in accordance with the error vector magnitude, until the EVM is reduced to a predetermined acceptable level; feeding the updated model back to a predistorter; and predistorting an incoming signal to generate said predistorted in . 5 coming signal; thereby allowing the implementation of an IQ feedback predistortion loop through the use of a sub-band receiver with an adjustable centre fre quency. For a better understanding of the present invention, reference will lo now be made, by way of example only, to the accompanying drawings in which: Figure 1 compares the characteristics of (ideal) linear and non-linear components; Figure 2 shows a block diagram of a PA arrangement with feed- for 15 ward linearization; Figure 3 shows a block diagram of a PA arrangement with adaptive analogue predistortion; Figure 4 shows a block diagram of a PA arrangement with adaptive digital predistortion; to Figure 5 shows a block diagram of a PA arrangement including compensating apparatus in accordance with the present invention; Figure 6 illustrates the signal waveforms formed at various compo nents of the PA arrangement; Figure 7 shows an example of a PA transfer function model used in 25 the simulated branch of the PA arrangement; Figure 8 shows a Non-Uniform Rotational B-Spline (NURB) curve (in bold) generated from a plurality of basis functions (dotted line);

Figure 9 illustrates power axis compression when converting from PA model to predistortion look-up table (PDLUT); Figure l O illustrates a corresponding power axis expansion when converting from PDLI IT to PA model; and 5 Figure 11 shows the architecture of an inverse PA transfer function component suitable for predistorting source data in accordance with up dated look-up table data.

Figures l to 4 are described above in the discussion of prior art.

A block diagram of a PA arrangement (500) with compensating ap o paratus is shown in Figure 5. The diagram shows two signal paths: the ac tual transmit path (502), including a PA (510) and a monitoring receiver (512); and a simulation path (504), comprising a software model (520) of the PA and of the monitoring receiver (522).

An initially un-predistorted signal (550) is transmitted via both paths 15 (502, 504). The last stage in each path is a digital filter (540, 540') that is applied to minimise the bandwidth difference between the two chains. The resulting signal (560, 560') is input into an error vector magnitude (EVM) calculator (530). An EVM measurement is made in order to quantify the difference between the two paths. EVM measurement is a conventional to technique specified in GSM standards documents.

Ignoring other transmit/receive imperfections, this "path EVM" (532) will be close to zero provided the PA model (520) in the software chain (504) is a good representation of the real PA (510). A predistortion algorithm is therefore applied, the algorithm seeking to adjust the PA 25 model to minimise path EVM. The digital filters placed before the EVM measurement allow substantially the same frequency band to pass in each of the two chains, thereby minimising the effects of earlier conversion stages upon EVM measurement. The hopping receiver (512), and a simu

lated model thereof (522), are arranged to have a wider bandwidth than these digital filters (540, 540'). The simulated model (522) of the down converter is an optional feature; it simply provides the option of modelling imperfections, where necessary. A further optional feature, shown in out 5 line in Figure 5, is a model (528) of an up-converter (508) . This model (528) can also be included to ensure that on the simulation path (504), the predistorted signal (552) arrives at the model PA (520) as it does for the real PA (S 10).

The PA model (520) is updated in accordance with the output of the lo EVM calculator (530). This updating of the PA model (520) is done itera tively in an attempt to improve the EVM measure (532) by observing the effects of adjustments to the gain and phase transfer functions of the PA model (520). This is a process that can be done "offline" - i.e. without in volving the PA for each and every adjustment - by recalculating the output 15 waveform of the simulation path (560) using the adjusted PA model (520) and therefore obtaining a new EVM measurement. This iterative adjust ment (540) of the PA model (520) can continue offline until a reasonable improvement is made. The inverse (536) of the PA model gain and phase transfer functions can then be used to update curves stored in a predistor o lion look-up-table (PDLUT) (506) and the process starts again. In Figure 5, the offline approximation loop (540) is picked out by thicker process ar rows. As the predistortion waveform (552) is applied to both paths (502, 504) in Figure 5 and as the algorithm is only concerned with producing a 25 PA model (520) that mimics observations of the real PA (510), the PDLUT update does not prevent the successive approximation loop (502, 532, 534, 538) from continuing as it did before.

The whole system illustrated in Figure 5 operates in the desired manner because the hopping receiver has a programmable centre band al-

lowing the use of methods normally only effective over a limited band width to be applied across a sufficiently large bandwidth? which in turn al 5 lows the use of IQ feedback techniques for predistortion.

In case of multiple sub-frequency bands, an EVM measure is calcu-

lated for each sub-band applied to the output of the PA, and the results are combined into a single metric, e.g. a mean EVM figure, in order to score the PA models ability to model the distortion over all applied subbands.

lo Figure 6 illustrates the signal waveforms formed at various compo nents of the PA arrangement (500). The waveforms show a snapshot of the signal spectrum in the frequency domain; the graphs having frequency on the abscissa and amplitude on the ordinate. Figure 6 suggests that the digital filtering (by the EVM minimising filters (540, 540')) must be such 15 that there is minimal difference between the remaining signal components before the EVM measurement. To that end, either: the digital filters must have a narrower bandwidth than the receiver filter (512b) and must have a corresponding fast roll-off to exclude signal components that differ due to different filtering levels; or the simulation path (504) must include a model so (522) of the receiver filter to minimise differences before the final digital filtering. The PA non-linearity is modelled with gain and phase curves as a function of instantaneous input power, as shown in Figure 7.

The PA model (520) in Figure 7 comprises: a first multiplier (720); a 25 phase look-up table [LUT] (730); a gain LUT (740); a phase delay unit (750); a second multiplier (760) and a third multiplier (770).

The first multiplier (720) operates to square the amplitude of an in coming signal (702) thereby generating a squared signal (704). The gain

LUT (740) generates a gain value (706), which corresponds to the squared value (704) applied. Similarly the phase LUT (730) generates a phase value (708) corresponding to the squared value (704). The phase value (708) is applied to the phase delay unit (750) to generate a phase adjusted 5 signal (710).

The phase adjusted signal (710) and the gain value (706) are multi-

plied together at the second multiplier (760). The output (712) of the sec-

ond multiplier is multiplied by the original incoming signal (702) at the third multiplier (770). A signal (714) representing the output of a PA is lo generated by the third multiplier (770).

As may be seen the key components of the PA model (520) are the gain and phase curves stored in the LUTs (740, 730).

Advantageously these curves for the PA model are adjustable in such a way that minor corrections can only be made to certain parts of the curve.

15 Furthermore the curves must also not be allowed to develop sharp bends.

The curves must rather provide fairly quick roll-off to model compression.

A known graphical method for creating suitable smooth adjustable curves is the Non-Uniform Rotational B-spline (NURB) technique. With NURBs, adjustments can be made without affecting other parts of the curve away to from the adjustment point. This is the method preferably adopted in the PA model. The use of NURBs in linearising non-linear behaviour is discussed at some length in UK patent application GB 0029420.7.

Figure 8 illustrates a typical NURB curve (804). The NURB curve (804) is constructed from a plurality of basis functions (806) (the light dot-

25 ted lines in Figure 8), which are scaled and summed to produce a smooth curve (804) (the bold line). The maxima of the basis functions (802) are commonly referred to as "handles" and value of these peaks, which relate to the scaling of the basis function, is the handle value. Increasing and de

creasing the value of the handles (802) that make up a given curve, pushes and pulls the curve (806) into the desired shape. In Figure 7 one such NURB curve is used for gain distortion and another NURB curve is used for phase distortion. The abscissa in both LUTs represents normalised lin 5 ear instantaneous power, which in turn is proportional to the squared am plitude of the incoming signal.

As there are a finite number of points in the NURB curves, values for instantaneous powers between points are calculated via linear interpo lation when applying the curves to data.

lo Each NURB handle value is adjusted by steps of a given size starting with the low instantaneous power end of one of the curves (the phase curve by default) through to the high instantaneous power end of the other curve.

The handle adjustments are as follows: firstly reduce the handle value by the step size until the measured 15 EVM increases in comparison with the previous handle position; next remove the last handle adjustment as it made the EVM worse; then increase the handle value by the step size until the measured EVM increases in comparison with the previous handle position; and finally remove the last handle adjustment as it made the EVM worse.

to Advantageously this process may be repeated from the concluding handle position with the step size successively halved upon each repetition to increase adjustment resolution. Once every handle on the gain and phase curves has been adjusted, the predistortion LUT is updated as explained above and the handle adjustment process is repeated. The initial step size 25 iS reduced with reducing EVM, with the following relationship: Si = SO en J

where s' is step size for iteration i and e, is the EVM for iteration i.

This has the effect that bigger (coarser) adjustments can be made for low reductions in EVM to quickly converge to the rough gain and phase curves required, while smaller (finer) adjustments are made when EVM is low and 5 tracking of long term changes is required.

Referring back to Figure 5, here the predistortion of the incoming signal is performed by means of a pair of look-up tables, again 'gain' and phase'. As was noted earlier, the inverse of the gain and phase transfer functions used to model the PA may also be used to update the values lo stored in the PDLUTs (thereby altering the predistortion curves).

Consider the equations for converting between the PA model and the PDLUT set out in the Table below. Here gain and phase are represented by G(P) and (Up) respectively, with P denoting input power. Equation (1) is used for compressing the power axis when converting from the PA model is to the PDLUT. Equation (2) is used for expanding the axis in the inverse operation of converting from PDLUT to PA model.

Converting from PA to PDLUT Converting from PDLUT to PA GPA(PPA)) | (1) PPA() = PLUTO)- GLUT(PLUT| (2)

GL UT() = 1/GpA (if) GPA (I) = 1/GLUT() AL UT (]) = -óPA (A) (PA (A) = óL UT (4) PDLUT transfer function: PA transfer function: y = x {CtUT(PLuT(x))- eXP(jó(PLUT(X)))} y = x {GPA (PPA (X))- eXP(jó(PPA (X)))} Table 1

In a preferred embodiment of the invention a combined predistortion and successive approximation algorithm updates the PDLUT directly, in-

stead of updating the PA model. The associated power axis expansion isconsidered a potential source of instability. In this embodiment, the PA 5 model is the inverse of the PDLUT.

As is illustrated in Figure 9, it is possible for the inverted gain re-

sponse (904) to curve back on itself when the power axis is compressed to convert to PDLUT. This "curving back" is due to the gain-to-power char-

acteristics for a PA (i.e. drop in gain for high input power, c.f. Figure 1).

lo Curving back would mean that the predistortion gain curve could not be used where it becomes multi-valued and essentially renders the predistorter useless. If, on the other hand, it is the PDLUT that is being manipulated and inverted for the PA model then the power axis is expanded as shown in 15 Figure 10. This expansion, coupled with the nature of the predistortion curve (i.e. an increase in gain at high power) , means that the PA model curve is always stretched out and can never curve back on itself. The dis-

advantage of applying an expansion is that this conversion has to be done on every approximation loop rather than only when the PDLUT needs up o dating and therefore requires more processing time.

For this reason the present invention envisages selection of the most appropriate target for manipulation: PDLUT or PA model.

Figure I 1 shows the architecture of an inverse PA transfer function used to predistort source data. The predistorter is similar to the PA model 25 with the exception that gain and phase do not operate from the same in-

stantaneous input power. The gain predistorter causes the input power to the PA to be changed in order to compensate for compression. A resealing unit is provided for this purpose. Hence the PA is being operated over a

different range of its transfer function. In compensating for phase distor tion, the phase predistorter must operate in this extended range of the trans fer function. To do this, the instantaneous power after gain predistortion is used to address the phase LUT. The scaling unit is a mean power resealer, s ensuring that the mean power remains constant and thereby preventing an increase in mean gain in the previous stage from causing an overflow at the input to the phase LUT.

In error vector magnitude measurement, constant time offset, phase rotation and gain difference are each represented as a series of vector trans o formations. EVM can therefore represent differences of scale, phase nor malisation, synchronization and offset delays. Non-integer time is consid ered as a movement along the path between two consecutive sampled vec tor points, and can be represented as a fraction of each path length making it a constant adjustment. Phase and gain are simple rotation and scaling 15 factors that can be applied to every symbol. Applying the vector transfor mations to the distorted data yields an individual vector error shown as: EX = Rx([Dx + [DX-l - Aft) À (1- A) where: Ex = error vector; to Rx = received vector; Dx = source data vector; r= time offset factor; 0= phase offset; and B= gain difference.

25 A Least Squares Minimisation is applied to the above definition of Vector Error in order to yield values of it, and,B that result in a mini mum Sum Squared Error over the data sequence.

Glossary of terms: (H)PA (High) Power Amplifier ADC Analogue-to-Digital Converter DAC Digital-to-Analogue Converter EDGE Enhanced Data-rate GSM Evolution EVM Error Vector Magnitude GSM Global System for Mobile Communications GPRS Global Packet Radio System IMD Intermodulation distortion, IMN when followed by a number N this refers to the Nth order intermodulation component(s) IQ In-phase / Quadrature LUT Look-Up Table NURB(S) Non-Uniform Rotational B-Spline PSD Power spectral density PDLUT Pre-Distortion Look-Up Table RF Radio frequency Rx Receiver Tx Transmitter UNITS Universal Mobile Telecommunications System

Claims (10)

CLAIMS:

1. A compensating apparatus for compensating for non-linear effects caused by one or more electronic components in a first signal path, the 5 compensating apparatus comprising: a predistortion device that operates upon incoming signals, generates a predistorted signal and feeds said predistorted signal into said one or more electronic components; a monitoring receiver that generates a first output signal, the receiver lo being provided in the first signal path and coupled to said one or more electronic components; a processing device that provides a second signal path, the process ing device receiving said predistorted signal, being arranged to generate a second output signal that simulates the non- linear effects of components in 15 the first signal path; and an error measurement device, which compares the first output signal and the second output signal and outputs an error value in dependence thereupon; wherein the error value is fed into the processing device, the proc o essing device calculates a simulation in accordance with the error value and the predistortion device is fed with updated simulation data by the proc essing device.

2. An apparatus according to Claim 1, wherein

25 the error value is calculated for each of a plurality of iterations and fed into the processing device, the processing device calculates an updated

simulation in accordance with the error value and the predistortion device is fed with updated simulation data by the processing device.

3. An apparatus according to Claim 1 or 2, wherein s the monitoring receiver is tuneable to a predetermined centre fre quency and receives signals over a restricted sub-band, the receiver being operated, during each iteration, at a plurality of centre frequencies, thereby providing sufficient coverage of the full signal bandwidth present in the .. mcommg signals.

4. An apparatus according to one of the preceding claims, comprising: a receiver modelling unit, for simulating non-linear effects occurring within the monitoring receiver.

15

5. A compensating apparatus for compensating for non-linear effects caused by one or more electronic components, the compensating apparatus . comprising: a sub-band receiver having an adjustable centre frequency; and an IQ predistortion feedback circuit; to wherein the feedback circuit iteratively samples the output of said one or more electronic components, calculates an appropriate predistortion factor, and applies said predistortion factor to incoming signals before said incoming signals are fed to said one or more electronic components; and wherein the sub-band receiver is operated, during a single iterative 25 loop of the feedback circuit, at a plurality of centre frequencies in order to

provide sufficient coverage of the full signal bandwidth present in the in coming signals.

6. An RF transmitter arrangement comprising: a compensating appara s tus as claimed in any of the preceding Claims; and one or more electronic components, said one or more electronic components including a power amplifier.

7. A method for compensating for non-linear effects caused by an elec o ironic component, the electronic component receiving a predistorted in coming signal and generating an output signal, the method comprising: providing a sampled signal, by sampling the output signal; inputting said sampled signal into a receiver, the receiver being ad justable to a predetermined sub-band centred upon acentre frequency; 15 selecting a centre frequency at which the receiver is to be operated; mixing the sampled signal with a mixing signal at a predetermined mlxmg frequency; filtering the output of the mixing step to allow signals in the prede termined sub-band to pass; to converting the filtered, sampled signal to generate a digital sample signal; digitally filtering the digital sample signal to generate a digitally fil tered sample signal that occupies a restricted range of frequencies; generating a model of the transfer characteristic of the electronic 25 component;

processing the input signal in accordance with the model, thereby generating a model signal that simulates the non-linear effects caused by the component; digitally filtering the model signal to generate a digitally filtered s model signal, which represents the transfer characteristic over substantially the same range of frequencies as occur in the digitally filtered sample sig-

nal; calculating an error valuebetween the digitally filtered sample signal and the filtered model signal and updating the model of the component; lo feeding the updated model back to a predistorter; and predistorting an incoming signal to generate said predistorted in comma signal.

8. Method according to claim 7, wherein Is an error vector magnitude between the digitally filtered sample sig nal and the filtered model signal is calculated iteratively and the model of the component in accordance with the error vector magnitude is updated, until the EVM is reduced to a predetermined acceptable level.

to

9. Method according to claim 7 or 8, wherein an IQ feedback predistortion loop is allowed to be implemented through the use of a subband receiver with an adjustable centre frequency.

10. An apparatus substantially as hereinbefore described with reference as to the accompanying drawings.