Signal correction techniques

[0001] The invention relates to methods of, and apparatus for, controlling distortion counteracting equipment such as a predistorter.

[0002] In a case where signal handling equipment distorts a signal upon which it operates, it is known to use a lineariser to reduce distortion in the output signal of the equipment. It is known to perform linearisation adaptively by monitoring errors in the output signal and using information on these errors to adjust the linearisation to reduce the errors as far as possible. Typically, predistorters are used to linearise amplifiers.

[0003] One aim of the invention is to provide improvements in the manner in which distortion counteracting equipment, such as a predistorter, is controlled.

[0004] According to one aspect, the invention provides apparatus for adapting distortion counteracting equipment, wherein said counteracting equipment employs a group of coefficients to adjust a consequential signal in order to ameliorate distortion in an output signal produced by signal handling equipment in response to an input signal, the apparatus comprising error measuring means for measuring errors in the output signal, modifying means for modifying measured errors to render them less dependent on the input signal's amplitude and correcting means for changing said coefficients by amounts dependent on modified errors.

[0005] The invention also consists in a method of adapting a distortion counteracting equipment, wherein said counteracting equipment employs a group of coefficients to adjust a consequential signal in order to ameliorate distortion in an output signal produced by signal handling equipment in response to an input signal, the method comprising measuring errors in the output signal, modifying measured errors to render them less dependent on the input signal's amplitude and changing said coefficients by amounts dependent on modified errors.

[0006] In one embodiment, the consequential signal is the input signal to the signal handling equipment and the counteracting equipment is a predistorter.

[0007] Thus, the invention improves the ability of distortion counteracting equipment to converge to the best values for the coefficients to optimise the elimination of distortion in the output signal. This is achieved by reducing the dependence of the changes to the coefficients on the input signal amplitude (which can vary greatly, e.g. in the case of CDMA signals), i.e. by reducing the dependence of the coefficients' loop gains on the input signal's amplitude. This means that the coefficients converge equally quickly.

[0008] In one embodiment, the coefficients are indexed by a function of the input signal's amplitude such that the input signal's amplitude selects the coefficient or coefficients that are used in the distortion amelioration process at any given time. Preferably, the coefficients are arranged in one or more look-up tables indexed by functions of the input signal's amplitude.

[0009] In one embodiment, the measured errors are indexed according to a function of the input signal's amplitude, i.e. the measured errors are tabulated against a function of the input signal's amplitude. For example, measured errors can be obtained for each of a series of values of the function of the input signal's amplitude. Where the coefficients are also indexed against the same function of the input signal's amplitude, it is possible to arrange the adaption process such that a measured error is used in the adaption of the coefficient sharing the same index value.

[0010] Where a function, f(a), of the input signal's amplitude, a, is used to index coefficients or errors, the function may be simple, such as f(a)=k.a, or more complex, such as f(a)=k.a² (here, k is an arbitrary constant). In the simplest case, f(a)=a.

[0011] In one embodiment, errors in a particular parameter are measured and these errors are used to adjust coefficients which act on the same signal parameter.

[0012] In one embodiment, the coefficients are segregated into sub-groups, each sub-group for altering a different parameter of the consequential signal. The errors used to adapt coefficients in a sub-group may be errors in the parameter to which the sub-group pertains. For example, where the coefficients fall into amplitude and phase sub-groups, the coefficients in the sub-groups are adapted on the basis of amplitude and phase errors respectively. Where the coefficients fall into sub-groups for orthogonal components, errors in each orthogonal component are used to adapt the coefficients for the respective component.

[0013] In one embodiment, the error values are derived over a period of time to reduce the effects of noise on the error values. For example, the error values could be accumulated and averaged over a period of time. The error values may be processed to remove the affects of noise, e.g. by filtering the errors or by fitting them to a curve. Similarly, after adaption, the coefficients may be processed to remove the affects of noise, e.g. by filtering the adapted coefficients or by fitting them to a curve.

[0014] Preferably, the modified errors are substantially independent of the input signal's amplitude.

[0015] In one embodiment, at least some measured errors are modified by dividing them by the input signal's amplitude to make them substantially independent of the input signal's amplitude. Where a measured error is divided by the input signal's amplitude and the measured errors are indexed by a function of the input signal's amplitude, the measured error can be divided by the amplitude value which indexes that measured error.

[0016] In one embodiment, the changes to the coefficients are proportional to modified errors. Preferably, the changes to the coefficients are fractions of the modified errors. Advantageously, the fractions can be changed to optimise the adaption rate of the coefficients. For example, the size of the fractions may depend on the size of the errors such that the smaller the error, the smaller the fraction and the larger the error, the larger the fraction.

[0017] The invention also consists in a system comprising distortion counteracting equipment for ameliorating signal distortion in signal handling equipment and comprising adaption apparatus as described above for adapting the counteracting equipment.

[0018] The invention has been stated above in terms of methods of adapting distortion counteracting equipment; the invention extends also to programs for performing such methods. These programs could be held in a suitable data store such as a disk or memory.

[0019] In a preferred embodiment, the signal handling equipment is equipment for amplifying the input signal to create the output signal, comprising one or more amplifiers.

[0020] In the preferred embodiments, the distortion counteracting equipment is a lineariser. In a particularly preferred embodiment, the lineariser is a predistorter and the coefficients are for predistorting the input signal to the signal handling equipment. In another embodiment, the lineariser is a feed-forward lineariser in which the input signal is sensed, modified by the coefficients and combined with the amplifier output to give a linearising effect.

[0026] FIG. 1 illustrates an arrangement for predistorting a radio frequency input signal RF₁ for an amplifier 10. The input signal RF₁ is predistorted in the digital domain using digital predistorter 12 to produce a digitally predistorted input signal S₃ for amplifier 10. A downconverter 14 reduces the frequency of the RF input signal RF₁ to a frequency supported by the sampling rate at which the predistorter 12 operates. The downconverted input signal is converted to the digital domain by analogue to digital converter (ADC) 16. The predistorted input signal S₃ is returned to the analogue domain by digital to analogue converter (DAC) 18 and upconverted to a desired frequency by upconverter 20 prior to being supplied as an input to amplifier 10.

[0027] The predistorter 12 can be adapted on the basis of feedback from the output RF₂ of amplifier 10. Discrepancies between the actual output RF₂ of amplifier 10 and its ideal output are used to adapt the predistortion of the amplifier input signal. The feedback signal is downconverted at downconverter 22 to a signal which is supported by the sampling rate of the digital aspect of the system, comprising predistorter 12 and control unit 24. The output of downconverter 22 is digitised by ADC 26 and supplied to control unit 24 as a feedback signal S₂. The control unit 24 uses the digitised feedback signal S₂ and the digitised input signal S₁ (from ADC 16) to adapt the predistortion.

[0028] If the original frequency of the input signal RF₁ is not incompatible with the sampling rate of predistorter 12, then downconverter 14 is not needed. If downconverter 14 is absent, then upconverter 20 is not required unless it is desired to shift the predistorted amplifier input signal to a frequency different from that of the original input signal RF₁. If the sampling rate of the digital aspect of the system supports the frequency of RF₂ then feedback downconverter 22 is not needed. The digitisation performed by ADCs 16 and 26 is performed in such a way as to preserve the amplitude and phase information in the signals being operated upon. For example, if the input to one of the ADCs is at baseband then the signal must be split into two paths with one path being in phase-quadrature to the second path with both paths then being digitised separately.

[0029] In this embodiment, the amplitude and phase of signal S₁ are modified in the predistorter 12 by the action of look-up tables (LUTs). This can be achieved by using a gain look-up table L_G and a phase look-up table L_P, each indexed by a function f of the amplitude A₁ of signal S₁. Thus, when signal S₁ has amplitude A₁, value L_G (f(A₁)) is retrieved from look-up table L_G, value L_P (f(A₁)) is retrieved from look-up table L_P and these values are then used to predistort the signal S₁ according to the following equation:
S₃=L_G(ƒ(A₁)).S₁.e^jLp(ƒ(A¹⁾⁾, where S₁=A₁.e^ja

[0030] Alternatively, predistorter 12 can contain an in-phase LUT L₁ and a quadrature phase LUT L_Q. Again, these look-up tables are indexed by a function f of the amplitude A₁ of signal S₁ in order to provide values L₁ (f(A₁)) and L_Q (f(A₁)), respectively. The retrieved LUT values are then used to predistort the input signal S₁ in quadrature format according to the following equation:
S₃=L₁(ƒ(A₁)).S₁+j.L_Q(ƒ(A₁)).S₁, where S₁=A₁.e^ja

[0031] The function f of the input signal amplitude A₁ that is used to index the look-up tables can take any one of a large number of forms. The best form for f in any particular case is dependent on the characteristics of the modulation on the input signal RF₁. For example, the two most obvious forms are f(A₁)=C.A₁ and f(A₁)=C.A₁² where C is an arbitrary constant which can be 1.

[0032] To adapt the predistortion in response to errors in the output of amplifier 10, the control unit 24 compares the signals S₁ and S₂ to generate updated LUT values which are transferred to the LUTs in the predistorter 12. Prior to comparison, the signal vectors S₁ and S₂ are, within control unit 24, time aligned (for example by digitally delaying S₁) so as to remove any relative delay difference between them and phase aligned (for example by digitally adjusting the phase of S₁ or S₂) to eliminate any phase offset between them. These alignment processes enable an accurate comparison of S₁ and S₂ to allow the error between them and its relationship to the input signal amplitude A₁ to be determined independently of the modulation frequency.

[0033] The ideal or target output signal vector S_2t of amplifier 10 is defined as a linear function of the input, i.e. S_2t=G.S₁ where G is a constant. An error signal vector or error vector S_e is defined as the difference between the measured output signal vector and the target output signal vector, i.e. S_e=S₂−S_2t. The error vector S_e is then described in terms of amplitude A_e and phase P_e components or in terms of two orthogonal vectors I_e and Q_e depending on whether the predistorter LUTs are in the amplitude and phase format or the quadrature format respectively. The relationship between the vectors S₂, S_2t and S_e is shown in the complex signal space diagram of FIG. 2, which also illustrates the components A_e, P_e, I_e and Q_e of S_e (note that I_e and Q_e are not aligned with the I and Q axes of signal space but rather I_e is parallel to S_2t).

[0034] Thus, as time progresses, various values of S_e (in the appropriate one of the A_e, P_e and I_e, Q_e formats) are recorded. Since signal S₁ has a time varying amplitude A₁, the recorded error signals S_e are obtained for various input signal amplitudes A₁. The error signals S_e obtained are tabulated against bins or ranges of the function of the input signal amplitude f(A₁) that is used to index the predistorter LUTs. Between predistorter updates, all error signals S_e falling within the same bin are accumulated and averaged to derive a mean error for that bin. Notionally therefore, the table created by tabulating the error signals S_e against f(A₁) is a table of mean errors indexed by f(A₁). The table of mean errors versus f(A₁) can be filtered or curve-fitted to remove the effects of noise. The table of mean errors is then used to adjust the predistorter LUT values.

[0035] $L_{G\left(n\right)}\left(f\left(A_1\right)\right)=L_{G\left(n-1\right)}\left(f\left(A_1\right)\right)-r·A_{e\left(n-1\right)}\left(f\left(A_1\right)\right)·\frac{1}{A_1}$
L_P(n)(ƒ(A₁))=L_P(n−1)(ƒ(A₁))−r.P_e(n−1)(ƒ(A₁))

[0036] Alternatively, where the predistorter operates with in-phase and quadrature-phase LUTs, the look-up table values are adjusted using the following equations:

[0037] $L_{I\left(n\right)}\left(f\left(A_1\right)\right)=L_{I\left(n-1\right)}\left(f\left(A_1\right)\right)-r·I_{e\left(n-1\right)}\left(f\left(A_1\right)\right)·\frac{1}{A_1}$ $L_{Q\left(n\right)}\left(f\left(A_1\right)\right)=L_{Q\left(n-1\right)}\left(f\left(A_1\right)\right)-r·Q_{e\left(n-1\right)}\left(f\left(A_1\right)\right)·\frac{1}{A_1}$

[0038] In the four preceding equations, n−1 and n denote that the current LUT values (n−1) are used to produce the new LUT values (n).

[0039] The two pairs of equations above each describe a feedback control loop with the subtracted terms representing the feedback. The loop gain for each entry in the LUTs is defined as the ratio of the change in the feedback term to the change in the LUT value, i.e.

[0040] $\frac{\Delta \mathrm{feedback}\mathrm{term}}{L_{x\left(n\right)}-L_{x\left(n-1\right)}}$
where x is G, P, I or Q as appropriate. In the case of each of the four feedback control loop equations above, it can be shown that the loop gain is approximately proportional to r and is independent of the input signal amplitude A₁ (provided that the effect of the amplifier non-linearity is ignored). The independence of the loop gain from A₁ is due to the term

[0041] $\frac{1}{A_1}$
which appears in the feedback equations except that for L_P(n). Using the above equations, the LUT values are updated for all values of the index f(A₁). The revised LUT values can be filtered or curve-fitted to remove the effects of noise if desired.

[0042] The control unit 24 is able to vary the size of the loop gain by varying the size of r in response to the magnitude of the error vector S_e. If the distortion is judged to be high then the loop gain is made relatively large by setting r to a relatively large value so that the LUT values converge quickly to a solution which minimises the distortion in signal S₂. On the other hand, if the distortion in S₂ is low, then the LUT values are already approximately correct and a relatively small loop gain is selected by setting r to a relatively small value so that the effects of system noise and spurious signals on the shape of the look-up tables are minimised. The preferred method for setting r is to use a mathematical function to generate a number based on the mean error table information. For example, a sum of squares calculation may be performed on the mean errors in the table. A large result implies a large amount of distortion at the amplifier output and hence that a large loop gain is required to arrive at a solution for the LUT values. A small result implies that there is only a small amount of distortion in the output of the amplifier and hence that a small loop gain is required, giving slower convergence of the LUT values.

[0043] FIG. 3 is a flowchart which further explains the process of updating predistorter LUT values in the case where in-phase and quadrature LUTs are used.

[0044] FIG. 4 is a flowchart further explaining the process of updating predistorter values in the case where amplitude and phase LUTs are used.

[0045] In the embodiments described above, the predistorters are vector predistorters which are capable of reducing both AM (amplitude modulation) to AM distortion and AM to PM (phase modulation) distortion. It will be apparent to the skilled person that the predistorter could be a scalar predistorter which counteracts only either AM to AM or AM to PM distortion by providing only a gain or phase LUT respectively with the result that only gain or phase errors need to be tabulated for subsequently adjusting the LUT values.