Method and circuit for pre-distorting an input signal for supply to an amplifier
The present invention relates to a method and circuit for pre-distorting an input signal for supply to an amplifier, and in particular to the pre-distortion of signals for use in a transmit power amplifier of a telecommunications system.
Recent advances in telecommunications standards and technology require the use of very linear amplifiers which can handle signals of wide bandwidth and with high peak- to-mean power ratios. In order to linearise such amplifiers techniques using feed-forward and pre- distortion are known. Development has focussed on pre-distortion because feed- forward amplifiers are complex and costly due to the extra RF components required.
Analogue pre-distortion has limitations with accuracy and also cannot adjust to cope with changes in amplifier characteristics over time (memory effects). Digital pre-distortion techniques have been developed for AM to AM compensation (for changes in amplitude introduced by the amplifier) and AM to PM compensation (for changes in phase introduced by the amplifier in relation to the amplitude of the signal). These can be implemented by look up tables or directly as polynomials and may be gain and phase or complex gain based. They typically assume that the amplifier can be characterised by an instantaneous non-linearity as a function of the envelope of the input signal and that the AM to AM function does not have any turning points.
If this is assumed, the amplifier can be described as a complex gain function of the input envelope. It has been shown that if the inverse of this function is known it can be used as a pre-distorter. Alternatively, it can be used to calculate the original signal if applied to the output of the amplifier. This use is known as a post distorter and allows calculation of coefficients of the function offline using captured blocks of input and sample data.
Recent techniques have aimed to take into account the history of the signal (memory effects) on the amplifier output. These are also modelled as functions of the signal envelope, but include samples of the signal at different times in the recent past. These models can account for effects such as power supply ripple, gate bias ripple, temperature
and many other time varying effects. Amplifier models including memory effect can use the same model structures as have been used with pre-distorters, for example memory polynomials and truncated volterra series. Post distortion algorithms have been tried, with limited success due to noise sensitivity issues.
It is an object of the present invention to provide a method for pre-distorting a signal which offers improved linearity of the output signal from an amplifier.
According to the present invention, there is provided a method of pre-distorting an input signal for supply to an amplifier, the method comprising: calculating a memoryless pre-distortion function; estimating an error signal; subtracting the error signal from the input signal to generate an error-compensated signal; applying the memoryless pre-distortion function to the error-compensated signal to generate the pre-distorted input signal.
By estimating the error signal, the linearity of the output can be improved.
Preferably, the method further comprises the step of calculating an amplifier model with memory, and wherein said step of estimating the error signal comprises: applying the memoryless pre-distortion function to the input signal; applying the output of the pre-distortion function to the amplifier model with memory; and subtracting the input signal from the output of the amplifier model to estimate the error signal.
By estimating the error in this way, the model can account for both memory effects and instantaneous AM to AM and AM to PM distortion in amplifiers. The use of a forward memory model according to the invention gives better performance than an approximate inverse model. The calculation of the model is less sensitive to receiver noise than inverse models because the input signal to the model is the input to the system and not a measured
signal. A further advantage is that there is no need to invert a memory model during calculations. Inverting a memory model can be computationally complex and is not always possible.
Preferably, the method further comprises periodically recalculating the memoryless pre- distortion function and/or the amplifier model with memory.
Preferably, a previously stored version of at least one of the pre-distortion function and amplifier model with memory is used. There is then no need to calculate these models and functions before the method can be used and any start up delay is reduced.
According to a second aspect of the invention, there is provided a method of amplifying a signal, the method comprising pre-distorting the signal according to the first aspect of the invention; amplifying the pre-distorted signal; measuring the amplified signal; comparing the measured amplified signal to predetermined target performance measurements; and recalculating at least one of the pre-distortion function and amplifier model if the measured amplified signal differs from the predetermined performance measurements by more than a predetermined amount.
This ensures that maximum efficiency is obtained, without wasting processing power recalculating the function and models when performance is within acceptable limits.
Preferably, the above mentioned first or second aspects of the invention is implemented by a DSP, microprocessor or FPGA.
According to a third aspect of the invention, there is provided an amplifier circuit comprising a pre-distortion circuit with an input connected to an input signal and an output connected to an amplifier, the pre-distortion circuit comprising:
calculation means for calculating and outputting a memoryless pre-distortion function corresponding to a memoryless model of the amplifier; estimation means for estimating an error and outputting an error signal; subtraction means having an input of the error signal and the input signal, for subtracting the error signal from the input signal and outputting an error-compensated signal; pre-distortion means having inputs of the error-compensated signal and the memoryless pre-distortion function, for applying the memoryless pre-distortion function to the error-compensated signal, thereby generating an output signal.
According to a fourth aspect of the present invention there is provided a transmit amplifier for a telecommunications system including an amplifier circuit according to the first aspect of the invention.
Embodiments of the invention will now be described, by way of example only, with reference to the following drawings in which: Figure 1 is a block diagram illustrating the calculation of a residual error by the method according to the present invention; Figure 2 is a block diagram illustrating the application of the residual error to a transmit amplifier; Figure 3 is an example of one embodiment according to the present invention; Figure 4 is another example of another embodiment according to the present invention; Figure 5 is a flow chart of the process of calculating the functions and models used in an embodiment of the present invention; Figure 6 is a flow chart of the process of calculating the functions and models used in another embodiment of the present invention; Figure 7 illustrates amplifier output without using the method of the present invention; and Figure 8 represents the output of an amplifier using pre-distortion according to the present invention.
Figure 1 shows the calculation of residual error for use in the method according to the present invention. The input signal Vi(t) is supplied to a memoryless model 10 which corresponds to the amplifier to which the signal is intended to be supplied. The output of this memoryless model 10 is then supplied to an amplifier model 20 which includes memory effects. The input signal Vi(t) is then subtracted from the output of the amplifier model with memory to give a signal Ve(t) which represents the residual error existing between the amplifier models.
It will be appreciated that the residual error could alternatively be estimated by subtracting the input signal Vi(t) from the output of the memoryless model 10. An amplifier model with memory is not then required.
Figure 2 shows how this error is used to improve the output of a conventional memoryless pre-distorter 50. The memoryless pre-distorter 50 in this embodiment is the same as the memoryless pre-distorter 10. However, different models may also be used if required. The subtraction element 40 subtracts the error signal Ve(t) from the input signal Vi(t) to produce a signal which is input to the memoryless pre-distorter 50. Thus, by subtracting the residual error which includes an amplifier model with memory, the distortion in the memoryless pre-distorter due to memory effects is reduced.
In this embodiment, the memoryless pre-distorters are modelled as complex look up tables, although alternatively polynomials of gain versus input or other methods which describe instantaneous AM to AM and AM to PM can also be used. Likewise, the amplifier model including memory effects 20 is implemented as a memory polynomial function or any other amplifier model including memory. However this may also be implemented in alternative embodiments as a look up table.
Figure 3 shows an embodiment of the present invention making use of the error feed- forward. The input signal is multiplied by two in multiplier 65 before the residual error is subtracted. This further improves the accuracy of the model.
Figure 4 illustrates a second embodiment of the present invention. An input signal is supplied to the real time pre-distorter 70 which uses the methods described above in relation to Figures 1 and 2. The output from the forward error correction pre-distorter is then supplied to a DAC which converts the signal to an analogue baseband signal. This is then up converted in up convertor 90 for supply to the amplifier 60. The output of the amplifier 60 is then supplied to an antenna 100.
A coupling element 110 samples the output of the amplifier and supplies it to a down convertor. The output of this down convertor 120 is then converted to a digital signal by ADC 130. This is then supplied as an input to coefficient calculation element 140, which is also supplied with the original signal as a reference. The coefficient calculation block
140 then periodically updates the model of the amplifier and updates the models contained in the forward error correction block 70, which can be an FPGA or DSP microprocessor. Processing in block 140, which can be a DSP, microprocessor or FPGA, is carried out offline.
Figure 5 shows how the coefficients are calculated in the embodiments of Figure 4. In step
S 10 the input signal is applied to the amplifier and the effect of the pre-distorter block 70 is bypassed. This can be achieved by loading appropriate coefficients which cause the models of the pre-distorter 70 to have no effect, so that it does not modify the signal which passes through.
At step S20 the output of the amplifier is measured and this data is supplied back to the microprocessor 140. Therefore, this measured data represents the effect of the amplifier on the input signal without pre-distortion applied.
In step S30 the microprocessor calculates new coefficients for the memoryless pre- distortion model. Iterative methods such as LMS (Least Mean Squares) or RLS (Recursive Least Squares) techniques on look up tables or polynomials can be used for the calculation.
It is also possible to use polynomial fitting techniques such as DLS (Discrete Least Squares).
In step S40 the model is then updated with these calculated coefficients. The output of the pre-distortion block 70 now takes into effect the calculated memoryless pre-distorter.
In step S50 a block of measured data is captured from the amplifier output to determine the effect of this pre-distorter on the signal output from the amplifier. In step S60 this captured data is then compared to target amplifier performance levels. The target amplifier performance levels can be expressed for example as a percentage distortion introduced by the amplifier, or other measures such as spectral purity.
If the performance of the system does not meet targets the method moves on to step S70 which recalculates the coefficients of the memoryless pre-distorter. The forward amplifier memory model is also recalculated in step S80. Finally, the memory models are updated with the new coefficients in step S90.
The final steps of this method, from step S50 onwards, are carried out continually during operation of the amplifier. Therefore, in the event that the amplifier model does not allow the amplifier to meet target performance, the models are recalculated until the amplifiers performance is within an acceptable limit. It also allows the amplifier model to adapt to changes in amplifier characteristics over time without carrying out unnecessary recalculations of the amplifier models when performance is acceptable.
Figure 6 illustrates an alternative method of controlling the microprocessor 140. In this method the pre-distorter coefficients are initially updated with a pre-stored value, rather than initially calculating the coefficients. The pre-stored value could be determined by the manufacturer at time of manufacturing and stored in a non- volatile memory, or alternatively it could be a stored copy of the last used model of the amplifier. Using the stored last calculated model of the amplifier allows the system to start up quickly without requiring a long initial adjustment and calculation period. The remaining step of the method in Figure 6 correspond to steps S50 to S90 described in relation to Figure 5.
The memoryless pre-distorter models can be calculated using blocks of input data compared with sampled blocks of data output from the amplifier 60. Iterative methods
such as LMS or RLS techniques on look up tables or polynomials can be used for the calculation. It is also possible to use polynomial fitting techniques such as DLS. The pre- distortion coefficients used in block 70 may be used in a different form than that calculated by microprocessor 140 if it is more appropriate or efficient to alter the coefficients into a form which is better suited in real time processing in block 70 than the offline processing in block 140.
The coefficients in the amplifier memory model 20 can be calculated iteratively using look up tables with a LMS or RLS method or by using singular value decomposition (SVD) for polynomial coefficients.
The performance of the present invention is illustrated in Figures 7 and 8. Figure 7 shows the output of amplifier 60 using a single carrier WCDMA signal with 8dB peak-to-mean without any pre-distortion applied.
Figure 8 shows the same signal with the pre-distortion according to the present invention applied. It can be seen that there is a significant improvement in the signal close to the carrier frequency. There is also significant improvement in spurious emissions.