GB2315379A

GB2315379A - Reducing peak to average amplitude ratio in communication apparatus

Info

Publication number: GB2315379A
Application number: GB9614588A
Authority: GB
Inventors: Anthony Peter Hulbert
Original assignee: Roke Manor Research Ltd
Current assignee: Roke Manor Research Ltd
Priority date: 1996-07-11
Filing date: 1996-07-11
Publication date: 1998-01-28
Anticipated expiration: 2016-07-11
Also published as: GB2315379B; GB9614588D0

Abstract

A method of reducing a peak to mean amplitude ratio of a signal at an output of a pulse shaping filter, hereinafter referred to as the post-filtered signal samples, is provided, the said method comprising the steps of generating test signal data samples in dependence upon a plurality of discrete time signal samples representative of the signal before passing through the pulse shaping filter hereinafter referred to as the pre-filtered signal samples, and scaling the pre-filtered signal samples in accordance with the test data samples, thereby constraining the post-filtered signal samples to a predetermined threshold.

Description

NON-LINEAR FILTER APPARATUS The present invention relates to signal processors which operate to reduce amplitudes of signals to a predetermined level, more especially to signal processors which operate to reduce a peak to mean amplitude ratio of non constant envelope signals produced after convolution with a pulse shaping filter.

Signals with non constant envelopes are produced when a linear modulating signal passes through a band limiting filter. The band limiting filter operates to restrict the bandwidth of the linear modulating signal to a predetermined range of frequencies. The band limiting filter may be required to ensure that the spectrum of the transmitted signal does not extend beyond a range of frequencies determined, for example, by a regulatory body or an operating requirement of a system within which the transmitter is embodied.

Within Code Division Multiple Access (CDMA) systems it is often a requirement for spread spectrum signals from a plurality of sources to be transmitted contemporaneously. To effect contemporaneous transmission, CDMA systems are provided with a means for summing the plurality of spread spectrum signals, thereby forming a composite signal. The composite signal is likely to have a non constant envelope exhibiting a substantial peak to amplitude mean ratio. To limit the frequency range to a predetermined bandwidth the composite signal may be subsequently passed through a band limiting filter further increasing the peak to mean amplitude ratio of the signal.

Signal transmitters are usually provided with power amplifiers which operate to amplify a modulated signal before feeding the modulated signal to an antenna which operates to propagate electromagnetic waves representative of the modulated signal. In a case where a signal transmitter operates to transmit a non constant envelope signal, the signal transmitter must be provided with a more expensive power amplifier, than if a modulated signal had a constant envelope with substantially the same mean power output. This is because an amplifier for a non constant envelope signal must produce a peak power output corresponding to a peak amplitude of the non constant envelope signal, which therefore requires that devices embodied within the amplifier are fabricated to accommodate the peak power output, increasing the expense of the amplifier. Furthermore an amplifier which operates to amplify non constant envelope signals operates during a substantial portion of time below the peak power level, as a result of which operation of the amplifier is considerably less efficient, than if the amplifier were to operate at substantially the mean power level. For these reasons it is desirable to reduce the peak to mean ratio of a non constant envelope signal prior to transmission, even if this results in some distortion and attendant degradation in sensitivity of a receiver provided for reception of the non constant envelope signals.

A reduction in a peak to mean ratio of a non constant envelope signal may be provided by passing a non constant envelope signals through a clipper after the signals have passed through the band limiting filter. The clipper operates to limit the amplitude of the signal. However the effect of such a clipper is that non linear affects associated with such clipping may cause the signal to extend beyond the predetermined range of frequencies previously imposed by the band limiting filter.

It is an object of the present invention to substantially reduce a peak to mean amplitude ratio of a non constant envelope signal.

According to the present invention there is provided a method of reducing a peak to mean amplitude ratio of a signal represented as a plurality of discrete time signal samples at an output of a pulse shaping filter hereinafter referred to as the postfiltered signal samples, the said method comprising the steps of; (i) generating test signal data samples in dependence upon a plurality of discrete time signal samples representative of the signal before passing through the pulse shaping filter hereinafter referred to as the pre-filtered signal samples, the post-filtered signal samples being representative of a result of convolving the pre-filtered signal samples with an impulse response of the pulse shaping filter, and (ii) scaling the pre-filtered signal samples in accordance with the test data samples, so that the post-filtered signal samples are substantially constrained to a predetermined threshold.

Step (i) of the method of reducing a peak to mean amplitude ratio of the said post-filtered signal, may comprise the steps of; (iii) generating test signal data samples by convolving the prefiltered signal with an impulse response appertaining to the impulse response of the pulse shaping filter, and the said method may further comprise the steps of; (iv) identifying which of the test signal data samples have a modulus which exceeds a predetermined threshold hereinafter referred to as inordinate samples; (v) generating a plurality of modifier coefficients in accordance with the inordinate samples and the said pre-filtered signal samples, which modifier coefficients when used to scale the said pre-filtered samples substantially reduces the amplitude of the inordinate samples to the predetermined threshold; (vi) arranging for the pre-filtered signal samples to be scaled with the modifier coefficients before being convolved with the impulse response of the pulse shaping filter, thereby reducing the peak to mean ratio of the said post-filtered signal.

The pre-filtered signal samples may be representative of a plurality of data symbols, and the impulse response of the pulse shaping filter may represent a duration corresponding to a plurality of symbols. Each symbol period of the impulse response may be represented by a plurality of impulse response (i.r.) coefficients.

Step (v) of the said method may comprise the steps of; (vii) calculating the modulus of each of a plurality of test signal data samples representative of a symbol, and assigning a complex sample corresponding to the test signal data sample with the greatest modulus to a complex variable L = L,+ JLQ, (viii) identifying whether the test signal data sample represented by the complex variable L is an inordinate signal sample, by comparing the modulus of L with a threshold k, and if the modulus of L is less than or equal to k, setting all modifier coefficients to unity, and if not, proceeding with steps (ix) to (xi), the signal sample corresponding to L being an inordinate signal sample, (ix) calculating complex excess data given by Eel = L,{1- k/lLI1 and EQ = LQ{1- k/lLl} where ILI is the modulus of L, and representative'of an amount by which the corresponding real and imaginary components of the inordinate signal sample L, must be reduced so that the modulus of L is reduced to that of the threshold k, (x) calculating the modifier coefficients in accordance with the pre-filtered signal samples, the impulse response of the pulse shaping filter, and the excess data, so that when the modifier coefficients scale the pre-filtered signal samples, the inordinate signal sample L is reduced by an amount corresponding to the excess data.

Step (x) of the said method wherein the modifier coefficients are calculated in accordance with an amount by which the plurality of pre-filtered signal samples must be reduced, in order to reduce the inordinate signal sample by the excess data, may comprise the steps of; (xi) comparing the polarity of each of the real and imaginary components of a plurality of contribution data samples to the polarity of the corresponding real and imaginary components of the excess data, each of which plurality of contribution data samples corresponds to one of the plurality of pre-filtered signal samples scaled by a corresponding impulse response coefficient of the impulse response of the pulse shaping filter, and (xii) if the polarity of the real component of the contribution data sample is opposite to the real component of the excess data, setting the corresponding value of the real modifier coefficient to unity, and (xiii) if the polarity of the real component of the contribution data sample is the same as the polarity of the real component of the excess data, calculating the corresponding real modifier coefficient in accordance with a combination of the corresponding real contribution data sample scaled by the sum of all real contribution data samples squared and the real component of the excess data El, and (xiv) if the polarity of the imaginary component of the contribution data sample is opposite to the imaginary component of the excess data, setting the corresponding value of the imaginary modifier coefficient to unity, and, (xv) if the polarity of an imaginary component of the contribution data sample is the same as the polarity of the imaginary component of the excess data, calculating the corresponding imaginary modifier coefficient in accordance with a combination of the corresponding imaginary contribution data sample scaled by the modulus of the sum of all imaginary contribution data samples squared and the imaginary component of the excess data Ea The plurality of impulse response coefficients representative of the impulse response of the pulse shaping filter may be representative of a duration corresponding to a plurality of symbols and may comprise two coefficients per symbol, and the test signal data samples representative of the output of the test filter may correspond to a single symbol, represented as two signal samples.

Step (x) of the said method may further comprise the steps of; (xvi) identifying whether the inordinate signal sample corresponds to an on symbol position, hereinafter referred to as an even sample, or whether the inordinate signal sample corresponds to an off-symbol sample position hereinafter referred to as an odd sample, (xvii) calculating the modifier coefficients corresponding to steps (x) to (xv), in accordance with a first process if the inordinate signal sample is an even sample, or calculating the modifier coefficients corresponding to steps (x) to (xv), in accordance with a second process if the inordinate sample is an odd sample.

The first process may operate to scale an odd number of pre-filtered signal samples (2s 1) with corresponding modifier coefficients numbered from -s+l to s -1 inclusive, in accordance with steps (x) to (xv) of the said method, wherein the modifier coefficients are calculated in accordance with an amount by which all (2s -1) pre-filtered signal samples must be reduced, in order to respectively reduce the real and imaginary components of the inordinate signal sample by the real and imaginary excess data components E1 and EQ., where s is an integer.

The second process may operate to scale an even number of pre-filtered signal samples (2s) with corresponding modifier coefficients numbered from -s+1 to s inclusive, in accordance with steps (x) to (xv) of the said method, wherein the modifier coefficients are calculated in accordance with an amount by which all (2s) pre-filtered signal samples must be reduced, in order to respectively reduce the real and imaginary components of the inordinate signal sample by the real and imaginary excess data components El and EQ Step (xiii) of the first process may comprise the steps of; calculating each of the real modifier coefficients for the i-th prefiltered signal sample in accordance with the following equation;

and step (xv) may comprise the step of calculating each of the imaginary modifier coefficients for the i-th pre-filtered signal sample in accordance with the following equation;

and step (xv) may comprise the step of assigning the modifier coefficients indexed, S, to unity, thereby causing them to be unmodified and so that c,, = 1 =1.

Step (xiii) of the second process may comprise the steps of; calculating each of the real modifier coefficients for the i-th prefiltered signal sample in accordance with the following equation;

and step (xv) of the second process may comprise the step of calculating each of the imaginary modifier coefficients for the i-th pre-filtered signal sample in accordance with the following equation;

The real coefficients of the impulse response of the pulse shaping filter may be designated aj, the coefficient ao corresponding to the centre coefficient. The real components of the pre-filtered signal samples may be designated Pi where i ranges from -s to s. The real component of the contribution data samples are designated a2i.pj, and the total contribution toLZis given by

The subscript of a is 2i and not i because the filter is two times over-sampled. Therefore alternate inputs to the filter are zero and contribute nothing to the output.

To modify the real components of the pre-filtered signal samples, so that the output of the pulse shaping filter is reduced by E" contribution data componentg,ip; is subtracted from each pre-filtered signal sample, which is achieved by multiplying the ith sample by the coefficient c, = (1 ). The component variables gl,i are hereinafter referred to as excess coefficients. By arranging

just enough may be removed from every sample so that L,is reduced by the right amount. However, if either the sign of i-th contribution data componenta2ipi is opposite to the sign of E, or if a2jpj =0, then empirically the excess coefficient g,; should be set to zero, so thatgl,i =0. This is reasonable since to do otherwise would require an increase in the absolute value of the sample; this could lead to the envelope exceeding the clipping threshold in unforeseen ways. This empirical observation may be represented mathematically by equation (5).

gl,i=γl,i#(Sgn(a2i.pi),Sgn(El)) (5)

1 x > 0 Sgn(x)=# 0 x=0 -1 x < 0 is the arithmetic sign function and,

1 i=j #(i,j)=# 0 i#j is the two argument Kronecker Delta function.

To calculate the other modifier coefficients, the corresponding values of gl,i may be scaled in proportion to the relative contribution of the i-th contribution data component a2iPi to the total. Thus, we let gl,i=gla2i.pi#(Sgn(a2i.pi),Sgn(El)) where gl is the normalising gain. Correspondingly the imaginary components of the modifier coefficients are calculated so as to reduce the imaginary component of the excess data Eq in accordance with the imaginary contribution data components of the pre-filtered signal samples a2iqi .

If the pre-filtered signal samples are representative of a non constant envelope signal, the method may further include the step of (x) pre-clipping the signal samples to the said predetermined threshold so that the amplitude of the pre-filtered signal samples is less than or substantially equal to the threshold.

According to an aspect of the present invention there is provided apparatus for reducing the peak to mean ratio of a signal represented as a plurality of discrete time signal samples at an output of a pulse shaping filter hereinafter referred to as the postfiltered signal samples, comprising a test filter means which operates to convolve pre-filtered signal samples representative of the signal before passing through the pulse shaping filter with an impulse response representative of the impulse response of the pulse shaping filter, a data processor being connected to the test filter means, which operates to generate a plurality of modifier coefficients, and a plurality of modification registers being connected to the data processor which operate to scale the pre-filtered signal samples with the said modifier coefficients, before being fed to the pulse shaping filter, thereby reducing the peak to mean ratio of the post filtered signal samples.

According to a second aspect of the present invention there is provided apparatus for reducing a peak to mean amplitude ratio of a signal represented as a plurality of discrete time signal samples at an output of a pulse shaping filter hereinafter referred to as the post-filtered signal samples, comprising a test filter means which operates to generate test data samples representative of a result of convolving pre-filtered signal samples representative of the signal before passing through the pulse shaping filter with an impulse response representative of the impulse response of the pulse shaping filter, a data processor being connected to the test filter means, which operates to generate a plurality of excess coefficients, and a plurality of modification registers being connected to the data processor and to the test filter means which operates to scale the test data samples with the said excess coefficients, thereby reducing the peak to mean ratio of the post filtered signal samples.

According to a third aspect of the present invention there is provided a receiver for detecting and recovering a transmitted spread spectrum signal which has been clipped in accordance with the method of reducing the peak to mean amplitude ratio of a signal as hereinbefore described, comprising a reference code generator which operates to generate a signal representative of a signal used to spread the spectrum of the transmitted signal, a signal processor which operates to generate a modified prefiltered signal in accordance with the method of reducing a peak to mean amplitude ratio of a signal as hereinbefore described, a radio frequency down conversion unit which operates to detect and down convert the transmitted signal to a baseband signal representative of the transmitted signal substantially at baseband frequencies and a plurality of correlators, each of which plurality of correlators are connected to the radio frequency down conversion unit and which correlators operate to correlate the baseband signal with a phase shifted version of the modified prefiltered signal, which phase shifted version corresponds to a possible phase shift introduced by a modulating signal, and a second data processor connected to each of the said correlators which second data processor operates to perform a maximum likelihood sequence estimation process, thereby providing a means for detecting the most likely sequence of modulating symbols corresponding to the transmitted signal.

One embodiment of the present invention will now be described by way of example only with reference to the accompanying drawings wherein, FIGURE 1 is a block circuit diagram of non-linear filter apparatus for processing a signal in accordance with the present invention, FIGURE 2 is a block circuit diagram of a pre-clipper, FIGURE 3 is an illustration of a graph of output envelope level for input envelope level for the pre-clipper shown in Figure 2, FIGURE 4 is a table of parameters used to simulate an example operation of the non-linear filter apparatus, FIGURE 5 is a set of discrete time signal diagrams which illustrate example signal diagrams produced by simulation of the operation of the non-linear filter apparatus, FIGURE 6 is an alternative embodiment of an apparatus for providing the method of reducing the peak to mean amplitude ratio in accordance with the present invention.

Although a spread spectrum signal generated in accordance with a CDMA system may have a constant envelope, when combined with a plurality of like modulated CDMA signals a composite signal resulting from the combination will possess a substantial peak to mean amplitude ratio. When this composite signal is passed through a pulse shaping filter, the peak to mean ratio of the resulting signal will have an even higher peak to mean ratio. An illustrative embodiment of the present invention which operates to reduce this peak to mean ratio may be seen in Figure 1. In Figure 1 three spread spectrum signal sources 1, 2, 3, generate complex baseband signals which are fed to two adders 4, 5. The adders 4, 5, operate respectively to sum the inphase and quadrature components of the complex spread spectrum signals fed from the CDMA signal sources 1, 2, 3 via conductors 7,9. The inphase I and quadrature Qsignals are fed to two inputs 8, 10, of a pre-clipper circuit 12. Two outputs of the pre-clipper circuit 14, 16, feed respectively I and Qsignals to first and second test filters 18, 20. The two test filters 18, 20, operate respectively to convolve the pre-clipped pre-filtered signal with an impulse response representative of the impulse response of the pulse shaping filters 62, 64. An output of each of the test filters 18, 20, is fed to an over-clip computer 24, via parallel conductors 26, 28.

Each of a plurality of conductors 30, respectively connect a corresponding output from the over-clip computer 24, to a first input of each pair of fours pairs of multipliers 32, 34, 36, 38, 40, 42, 44, 46.

The shift register 47, 48, 50, 52, 54, 56, 58, 60 and first 32, 34, second 36, 38, and third 46, 42, and fourth multiplier pairs 46, together comprise the modification registers 80.

The I and Qoutputs 14, 16, from the pre-clipper 12, are respectively connected to two delay shift registers 47, 48 which operate to introduce a delay into a discrete time signal, represented as z-n. An output of each shift register is respectively fed to a second input of the multiplier pair 32, 34.

Respective outputs of the multiplier pair 32, 34, are respectively connected to a second pair of output shift registers 50, 52 which thereafter feed a second pair of multipliers 36, 38. Respective outputs of the second multipliers 36, 38 are connected to a third pair of shift registers 54, 56, respective outputs of which are connected to a third pair of multipliers 40, 42. Respective outputs of the multipliers 40, 42, are connected to a fourth pair of shift registers 58, 60, an output of which is connected to a second input of the fourth pair of multipliers 44, 46. Respective outputs of the multipliers 44, 46, are fed to first and second pulse shaping signal filters 62, 64, which operate to respectively convolve real and imaginary signal samples received from multipliers 44, 46.

Respective outputs of the first and second filters 62, 64, are fed to a fifth pair of multipliers 66, 68, a second input of which fifth pair of multipliers 66, 68 is fed from a local oscillator 70.

Respective outputs of the pair of multipliers 66, 68, is fed to an RF up converter and RF output circuit via a conductor 80.

The I and Qcomponents of the complex composite signal formed from the sources of spread spectrum signals 1, 2, 3 via the adders 4, 5, are respectively fed to I and Inputs of the preclipper 12. The pre-clipper 12, operates to pre-clip the composite signal so as to reduce the peak to mean ratio of the signal by providing that the amplitude of the composite signal does not exceed a predetermined value. The pre-clipper 12 will be described in more detail with reference to Figure 2 in the subsequent paragraphs. The output of the pre-clipper circuit 12, represents I and Qvalues of a complex signal which are respectively fed to test filters 18, 20. The test filters 18, 20, operate to convolve the pre-clipped signal with a representation of the impulse response of the pulse shaping filters 62, 64. The result of the convolution is fed to the overclip computer 24, via the parallel bank of conductors 26, 28. The over-clipped computer 24, operates to generate a number of modifier coefficients which are applied to the first input of first, second, third and fourth multiplier pairs 32, 34, 36, 38, 40, 42, 44, 46.

The modifier coefficients operate to scale the pre-clipped signal samples of the composite signal so that when the scaled signal is convolved with pulse shaping filters 62, 64, the amplitude of the composite signal remains below a predetermined threshold.

Calculation of the modifier coefficients will be described in subsequent paragraphs. In operation therefore, signal samples output from the pre-clipper 12, are also fed to the first input of each of the pair of delay shift registers 47, 48 for respectively the I and Qcomponents of the complex signal samples. The first shift registers 47, 48, operate to reproduce the delay in passing the pre-filtered signal through the test filters 18, 20, and calculating the modified coefficients such that causality is maintained in the processing of the signal. Thereafter the bank of parallel multipliers and shift registers operate to multiply the pre-clipped composite signal samples by the modified coefficients calculated by the over-clip computer 24. Following multiplication by the modified coefficients the I and Qcomponents are respectively fed to the input of the pulse shaping filters 62, 64, which convolve the impulse response of the pulse shaping filters with the modified pre-clipped composite signal which is thereafter fed to the multipliers 66,68 which is up converted in combination with the local oscillator 70, and phase shifter 72, and thereafter fed to RF amplifier and further components via the conductor 82.

An embodiment of the pre-clipper 12 may be seen in Figure 2. In Figure 2, I and Qcomponents are fed to the pre-clipper 12 via conductors 100 and 102. The conductors 100 and 102 are connected respectively to a first input of a pair of multipliers 104 and 106. The I and Qconductors 100 and 102 are also connected to an amplitude function 108, which operates to produce a signal representative of the amplitude or envelope of the complex signal sample represented by the I and Qcomponents, in accordance with the expression, 412 + Q2 . The amplitude of the signal sample is thereafter fed to a threshold detector 110, which operates to determine whether the amplitude calculated by the amplitude calculator 108, is above a threshold determined and indicated as the value k. The threshold detector 110, operates to pass the value k, or the envelope 7my2, to the scaling means 112, depending on which is the greater. Thereafter if the signal sample has an amplitude which is greater than the value k the scaling means 112, operates to feed a signal to a second input of the pair of multipliers 104, 106, which signal is provided so as to scale the I and Signal samples by a factor k ensuring that the amplitude of the signal samples remains below the threshold k. Thus, if the envelope is lower than k, the scaling factor is unity. If the amplitude of the signal sample is greater than k, the signal is scaled such that the envelope becomes equal to k.

The operation of the pre-clipper circuit 12 is illustrated by the graph shown in Figure 3. In Figure 3 the line 120 represents the output function of the pre-clipper circuit 12, for the composite signal input envelope level. As can be seen where the input envelope level is below a value of k then the amplitude of the signal remains unchanged, however where the input amplitude level exceeds a value of k the output envelope is clipped to a value of k, as illustrated by the flat portion of the line 122.

The illustrative embodiment of the present invention uses the pre-clipper 12 since this is appropriate in the context of the addition of several signals together (as may arise in the base station transmitter of either a CDMA cellular mobile or Wireless Local Loop system). However, in the case of a single signal the pre-clipper 12, becomes unnecessary since, prior to filtering, the signal is constant envelope in any case.

If a pre-clipped signal is applied to a lowpass filter, ringing in the filter will result, and on occasions, the signal level at the filter output may exceed the clipping threshold. If clipping were applied without regard to the effect of a band limiting or low pass filter, an expansion of the signal bandwidth would result, due to discontinuities in slope at the clipping point. To obtain a clipped signal at the output of the pulse shaping filter, the output of a test filter operating at several samples per symbol (or chip in the case of direct sequence spread spectrum) is examined and used to generate a set of modifier coefficients which operate to scale the pre-clipped signal. The non-linear filter apparatus depicted in Figure 1, therefore operates to reduce the peak to mean

4. Passing the modified pre-filter signal sample stream through the pulse shaping filter to produce the required clipped filtered output.

The apparent non-causality between steps 2 and 3, is removed by inserting the delay shift registers 47, 48, into the path of the pre-filtered signal samples at the input to the modifier registers, which have the effect of delaying the discrete time prefiltered signal by n-samples.

Operation of the overclip computer 24, will now be described for a case where the test filters 18, 20, operate at two samples per symbol (chip) by way of example only, since other sample rates are also possible. Furthermore the embodiment of the invention will be described with reference to a signal modulated in accordance with M- Phase Shift Keying (M-PSK), primarily with M=2 and M=4, which is BPSK and QPSK. The description is most easily understood in terms of QPSK (since the actual data to symbol mapping is not considered), however, all possible modulation schemes are implied, including DPSK. In the case of BPSK, the description can be extended by setting the input signal on the imaginary ( Q) channel to zero and simplifying the resultant expressions and their corresponding implementation.

The pre-clipper 12, operating as illustrated in Figure 2, first pre-processes the pre-filtered signal samples, which are subsequently fed to the test filters 18, 20, operating at two samples per symbol (chip). The outputs of the test filters 18, 20, are examined to determine which signal sample have an envelope which exceeds the clipping threshold, k. Signal samples which cause this envelope to exceed the threshold are herein after referred to as inordinate signal samples. When this happens, these inordinate signal samples are scaled by the modifier coefficients before being applied to the pulse shaping filters. The delay, of n samples corresponds to the pre-cursor samples of the test filter response up to half the filter length minus two samples.

In this way four samples corresponding to a symbol produced at an output of the test filter, which comprises two complex signal samples in this example embodiment are multiplied by modifier coefficients corresponding to those samples which will become multiplied by the maximum impulse response coefficients of the first and second pulse shaping filters 62, 6.4 to produce the output symbol, as was correspondingly just previously produced by the test filters.

In the general case, the number of complex samples subjected to overclipping may be any even number, 2s, from 2 upwards. There is unlikely to be any benefit from increasing s beyond 2. The sample positions range from-s +1 to s where the higher numbered sample is the most recently generated.

The delay shift registers 47, 48, 50, 52, 54, 56, 58, 60, associated with the multipliers 32, 34, 36, 38, 40, 42, 44, 46, are clocked once per symbol (chip),whereas the test filters 18, 20, are clocked twice per symbol (chip). Thus the overclip computer 24, operates to produce one set of output coefficients for each symbol produced at the output of the test filters 18, 20, which in this case comprises two complex samples.

Calculation of the modifier coefficients is provided by either of two processes, which operate within the overclip computer 24, in dependence upon whether either of the two complex samples appertaining to a symbol produced at the output of the test filters 18, 20, corresponds to either odd or even samples. An even signal sample is defined as being a signal sample which corresponds approximately to the 'on symbol (chip)' position, whereas an odd signal sample corresponds approximately to an 'off (or halfway between) symbol (chip)' position. The odd and even cases are dealt with in slightly different ways, and are therefore processed by a first of the two or a second of the two processes respectively.

The steps involved in the operation of the overclip computer 24 are presented as follows: 1) After two cycles of a clock of the test filter 18, 20, the amplitude of both the odd and even signal samples is computed.

The larger, L, of these samples is determined. If the larger is the odd sample, the second process is used to compute the modifier coefficients. If the larger is the even sample, the first process is used to compute the modifier coefficients.

2) If ILI < k all modifier coefficients c1 and CQ^ are set to 1 where ILI is the amplitude of L, and the processing is complete for this sample position.

3) If ILI > k real and imaginary excess data components are computed which is representative of an amount by which the corresponding real and imaginary components of L must be reduced to make ILI =k. Thus, if we let L = L, + jLQ, the real and imaginary excess data are given by El = L1(1-k/ILI} and EQ = LQ{l -k/lL4} respectively. We now need to reduce the relevant inputs to the pulse shaping filters in such a way as to reduce L by the excess data. The delays in the paths to the pulse shaping filters are arranged to be such that signal samples, which when shifted into the position in which they are multiplied by the centre taps of the pulse shaping filters, will produce the test filter outputs which have just been generated. Note that any amount of latency in the computations in the overclip computer 24 (eg due to pipelined computations) can be compensated by increasing the delay in the paths to the modification registers 80.

4) If the greater modulus was found on the even signal sample, the excess data, in the first process is applied, which operates to remove excess data by modifying an odd number of samples (2s -1) numbered from -s +1 to s-l inclusive. The middle (zero referenced) of the 2s-1 samples will be the sample which, when multiplied by the largest impulse response coefficient will produce the output of interest in the test filter (i.e. in the test filter, the same value will have just been multiplied by the largest coefficient). The other samples will be multiplied by the impulse response coefficients symmetrically disposed about the centre coefficient. For the real channel, we designate these samples, where i ranges from -s to s. The total contribution of these J-1 samples to Llis given by #a2i.pi where ai are the impulse i=-s+1 response coefficients and a0 is the centre coefficient. The product a2,p, are hereinafter known as contribution data, since these are formed in the convolution sum and contribute to each output of the test filters. The reason why the subscript to a is 2i and not i is because the impulse response of the test filters 18, 20, and pulse shaping filters 62, 64, is two times oversampled so alternate inputs to the filter are zero and contribute nothing to the output.

We wish to modify the real inputs to the real signal filter so that its output is reduced by E,. Conceptually, this can be achieved by subtracting an amount from each of these samples. Supposing we subtract gl gl,ipl from the ith sample. This is achieved by multiplying it by the modifier coefficient cl,i=(1-gl,i). Then if we can arrange s-1 that #a2i.gl,i.pi=El we still have removed just enough from every i=-s+1 sample so that L1 is reduced by just the right amount. If either the sign of a2iPi is opposite to the sign of E, or if azipi =0 we empirically see that we should set g,i =0. This is reasonable since to do otherwise would require an increase in the absolute value of the sample, this could lead to the envelope exceeding the clipping threshold in unforeseen ways. Thus we can define gl,i=γl,i#(Sgn(a2i.pi).Sgn(El)). Where Sgn(x) and #(i,j) are defined in equation (5).

The remaining values of the excess coefficients g, are calculated by scaling each of the values of g,,1 in proportion to the relative sample of the contribution data a2ipi, to the total. Thus, we let gl,i=gla2i.pi#(Sgn(a2i.pi),Sgn(El)) where g, is the normalising gain, then we have: s-1 #gla2l.pi#(Sgn(a2i.pi),Sgn(El))=El i=-s+1 SO

which leads to:

and finally to equation (1), which is reproduced as follows:

and similarly for the Qchannel:

In this case, an odd number of samples is modified, and for this reason the terms indexed, s, are unmodified and so the corresponding modifier coefficients are set to unity;czS =1 and C =1 If the greater modulus was found on the odd signal sample, then the excess data is removed by operation of the second process. In the case of the second process all2s samples, numbered from -s +1 to s, are processed. The middle two of the 2s samples will be those which, when multiplied by the two coefficients directly either side of the largest impulse response coefficient, will contribute to the output of interest in the test filters 18, 20. Consider the real channel only at this stage. The total contribution of the contribution data samples to the output given by

where the ai and p are as defined earlier.

Using identical working to that given previously we obtain the modifier coefficients for the relevant samples as according to equation (3) for the real or I-samples and equation (4) for the imaginary or Qsamples:

The above general case presented for the illustrative embodiment, implies considerable complexity. However simplicity of the operation of the example embodiment can be seen when we take the case of s=1. In the case that the even sample is the El larger, if Sgn(p0)=Sgn(El) then cl,0=1-, otherwise cl.0=1 and p0.a0 similarly for the imaginary channel. In the case where Sgn(p0)# Sgn(E,) it is not possible to remove the excess. In the case that the odd sample is the larger, if Sgn(p0)= Sgn(E,) and Sgn(p1)=Sgn(El) then Elp0 Elp1 cl.0=1- and cl.1=1a1(p0+p1) a1(p0+p1) If Sgn(pO) = Sgn(E1) and Sgn(p1) Sgn(E) then E1 cl.0=1- and cl.1=1 p0-a1 If Sgn(p0)#Sgn(El) and Sgn(p1)=Sgn(El) then c1,0 =1 and c1,1 = E1 p1.a1 If Sgn(p0)#Sgn(El) and Sgn(p1)#Sgn(El) then cl.0 = 1 and c,l = The forms for the imaginary channel are equivalent. In the last case above it is not possible to remove the excess data, E, by modifying signal samples which are available for modification.

Since the signal samples which are available for modification are those which contribute most to inordinate samples the only condition under which the excess data cannot be removed, is in a situation where the dominant cause of clipping arises on the other of the two phases. Thus, if the condition in which we cannot remove the excess data, E" arises this can only be in the case where |EQI IE I Eli and vice versa. In this case, failure to cancel the excess data will result in the envelope exceeding the clipping threshold, k, but by a negligible amount. The likelihood of this happening diminishes with increasing s.

A simplification of the general case of the illustrative embodiment of the present invention may be provided by constraining the process for reducing the peak to mean ratio so that modifier coefficients are generated in accordance with the amplitude of the odd or off-chip signal sample only. In this case the process for calculating the modifier coefficients would be reduced to the second process only, and in accordance with the operation of the process for reducing the peak to mean ratio hereinbefore described, the second process would only operate to generate modifier coefficients where the odd signal sample is an inordinate signal sample. This simplification might lead to some samples with amplitudes close to or substantially above the clipping threshold, however this is unlikely because the scaling operation of the modification coefficients would also substantially reduce the even signal samples.

As an illustration of the operation of the example embodiment of the method and apparatus for reducing a peak to mean amplitude ratio of a signal according to Figures 1, 2 and 3, an illustrative example of the effect of the clipping process is provided, as follows, with reference to a simulation analysis.

Parameters for the clipping process incorporated into the simulation analysis are provided in Figure 4, with the result of the simulation analysis illustrated by a set of discrete time signal diagrams presented in graphical form in Figure 5. Figure 5 illustrates the progression of operations through the non-linear filter apparatus. In Figure 5, a first timing diagram (a) is representative of the impulse response of the test filters 18, 20. A discrete time sample diagram (b) is representative of the contents of the test filter register components which correspond to contribution data components and a timing diagram (c) is representative of an inordinate signal sample, since this sample has an amplitude which is greater than the predetermined threshold. For the timing diagram (c) a timing diagram (d) represents the corresponding data excess value. In timing diagram (e) the contribution data associated with the excess data shown in timing diagram (d) are shown for which an amount to be subtracted from this amplitude is presented in timing diagram (f).

In accordance with timing diagram (f) timing diagram (g) presents a set of modifier coefficients which operate to reduce the excess data of the inordinate signal sample shown in (c) to a level which is below that of the clipping threshold. In timing diagram (h) a result of multiplying the pre-clipped signal samples with the modified coefficients is presented which discrete signal samples presented in timing diagram (f) are subsequently convolved with the impulse response of the pulse shaping filters 66, 68. The result of the convolution on the inordinate signal sample presented in timing diagram (c) is presented in timing diagram (e) wherein the amplitude has been reduced to unity, unity being indicative of the threshold of the desired maximum amplitude of the signal.

Discrete time signal diagrams presented in Figure 5, show signal samples representative of the values corresponding to shift registers with signals entering at the left, therefore the time axes are reversed. Thus the oldest samples are shown on the right and the newest on the left. This convention is used in each of the diagrams and is highlighted by the numbering of the tap weights in Figure 5(a). Examining Figure 5(b) we see that alternate samples are zero. This is because the filter operates at two samples per symbol (chip), thus interpolating between samples.

For the timing shown, we see that the sample corresponding to the zero indexed tap weight is zero valued. Thus the timing shown corresponds to generating the interpolated filter signal sample corresponding to an "odd" or 'off (or halfway between) symbol (chip)' position as described earlier.

Figure 5(b) shows the real, imaginary and modulus values in a single display according to the key given. Because the I and Q signal amplitudes are obtained from adding together some thirty values, each of which are +0.1, the output is generally a multiple of 0.2. However, the output corresponding to position -1 on the filter taps has modified values and a modulus exactly equal to 1.0 indicating that the pre-clipper has operated correctly.

Figure 5(c) shows the signal samples (real, imaginary and modulus) generated by the test filter for the current set of samples in the shift register. It maybe seen in this case that the amplitude or envelope of the output sample exceeds the clipping threshold.

Figure 5(d) shows the real and imaginary components of the excess data corresponding to the inordinate signal sample output by the test filter. Thus, both the real and imaginary components must be scaled by the factor 1/1.32. This can be viewed as subtracting (1 1/1.32) times the real value, from the real value and similarly for the imaginary value. Thus the amounts for subtraction, i.e. the Excess data are given by: E, 1.32(1 1/1. 32) 0.32 and.F = 055(l - l/l.32) = 0.0134 as shown in Figure 5(d). Note that the original computations were performed at high precision, but have been rounded for display purposes, which may lead to some minor discrepancies in the calculations. Note, also that in many cases it is difficult to see the imaginary component on the graphs because it is so small, but every display of a real value is actually accompanied by an associated imaginary value. Having computed the excess data we now wish to determine the modifier coefficients to feed into the modification registers 80. In this example s=1 we are therefore dealing with an odd output case, wherein there are two complex samples for modification in the modification registers. These correspond to signal samples at positions 1 and -1 in the test filter discrete time diagram of Figure 5(a) although the samples which will actually be modified (which currently should have identical values) are stored in the modification registers. We examine these samples to see how much they contribute to the overall output on both the real and imaginary channels. This is done by multiplying the sample amplitudes by the corresponding tap weights which, in this case have the value 0.629. These multiplications are performed (which can be confirmed by examining the samples at positions 1 and -1 in Figure 5(b)) and shown in Figure 5(e)). One might expect the contribution for the imaginary sample at position 1 to be -0.4XO.629=-0.252. However, since this contribution has opposite sign to the filter output it is not considered a useful contribution and is therefore set to zero, or, in other words, it would be necessary to increase this sample in order to reduce the excess data in order to avoid a situation where a post filtered signal sample which has an amplitude below the clipping threshold, may subsequently exceed the clipping threshold. The amount of these samples which must be removed in order to remove the real and imaginary components of the excess data can now be determined. For the real case it is easy to see that the amounts are:gI,1=EI.0.8 =0.296 and gI,0 = E1.0.857=0.318 as 0.629(0 +0.8 + 0.857) 0.629(0.8 + 0.857) seen on Figure 5(e). We can then confirm that E1 =(O.296x0.504)+(0.3l8x0.54) where the first terms in the multipliers are the excess coefficients gl.i and the second terms are the contributions as shown in Figure 5(e). Also, we see that then, have the same ratio as the contributions. For the imaginary case we only have one non zero contribution, so gQ, =1 and EQ gQ,0= =0.413 as seen on Figure 5(e). Here the 0.629x0.514 denominator of the expression for forgQ0 is equal to the contribution of that sample to the output.

As indicated in Figure 5(g), the modifier coefficients are computed by the simple equations:- cl,i=1-gl,i and cQ,i=1-gQ,i.

We now apply these modifier coefficients to the multipliers in the modification registers 80. Figure 5(h) shows that the samples older than those in the modification registers have already entered the shift register within the pulse shaping filter. The samples which are newer than those in the modification registers are contained in the initial delay circuit (labelledz~" in Figure 1).

Thus, being scaled by operation of the multipliers in the modification registers, the resulting signal samples will appear as shown in Figure 5(h). Only the centre two complex samples are different from the samples as they appear in Figure 5(b).

After the samples shown in Figure 5(h) have been shifted in time so that they are convolved with the pulse shaping filter in corresponding positions to those shown for the test filter impulse response, the signal samples will appear as shown in Figure 5(i).

Here, we see that the modulus of the inordinate signal sample has been reduced to exactly 1.0 (although, the modulus computed from the real and imaginary components is about 0.9997, the true computation is exact to many significant figures).

An alternative embodiment of the present invention may be provided by constraining every modification coefficient on the real channel to be the same as the corresponding modification coefficient on the imaginary channel. This avoids the generation of incidental phase modulation as a result of the overclipping.

This can be done in at least two ways, which are presented as follows The first of the two ways is to set the modifier coefficient for both real and imaginary channels to be equal to the lower of the coefficients over the real and imaginary channels. This guarantees that the clipping threshold will not be exceeded at the expense of clipping the signal envelope below the clipping threshold (ie greater clipping than is strictly necessary) The second of the two ways, is to solve the joint equations for the real and imaginary channels subject to the condition that the modifier coefficients are the same on both the real and imaginary channels. This gives an improved performance but is likely to be prohibitively complex.

An alternative embodiment of an apparatus for performing the clipping function in accordance with the present invention is presented in Figure 6. In Figure 6 parts also appearing in Figure 1 bear identical numerical designations. Figure 6 differs from Figure 1 in that the over-clip computer has been replaced by an alternative over-clip computer labelled 'over-clip computer 2', 200 and the outputs from the test filters 18, 20 for respectively the I and Qsignal samples are fed to a pair of shift registers 202, 204 by conductors 206, 208 which operate to delay the signal presented at the output of the test filters 18, 20, by a factor Z-n, The outputs from the shift registers are fed to a second pair of filters 210, 212 which filters are excess filters which operate respectively to filter the I and Qcomponents received from the test filters 18, 20, via the pair of delay shift registers 202, 204.

The outputs from the excess filters 210, 212 are fed respectively to one input of two subtracting amplifiers 214, 216, which operates to subtract respectively the I and Qcomponents at the output of the excess filters from that produced from the output of the test filters fed from the delay shift registers 202, 204.

Respective outputs from the pair of subtracting amplifiers are fed to the inputs of the pair of multipliers 74, 76 thereafter to be multiplied by the local oscillator frequency in accordance with upconversion of the processed signal for transmission of the signals.

Figure 6 illustrates an alternative embodiment of the non linear filter which operates to provide the process of reducing a peak to mean amplitude ratio of a signal. In this case the pulse shaping filters 70, 72 have been replaced by Excess Filters 210, 212 and the overclip computer 24, has been replaced with the alternative overclip computer 200. The difference here is that the alternative overclip computer 200, operates to generate excess coefficients corresponding to the aforementioned variables, g,i and gQ j, instead of the modifier coefficients c,i and Cm,1. If there is no clipping, the modifier coefficients are set to zero. The excess filters 210, 212, operate to remove excess signal component from a ringing signal produced at the output of the test filters. By subtracting the filtered excess signal from this ringing filtered signal, and compensating for the additional delay in the excess filter path a clipped band limited signal is generated for up conversion with characteristics substantially the same as that generated by Figure 1.

The alternative embodiment of the present invention as hereinbefore described, is provided with advantages over the earlier embodiment described with reference to Figure 1, in terms of complexity in that there is slightly less computation required in generating the excess coefficients g, i and gQ j than in generating modifier coefficients c1, and Cm,1. Furthermore since most of the excess coefficients will be zero, much of the computation in the excess filters 210, 212, will be trivial, resulting in reduced power consumption. It would also be possible to examine the contents of the shift register in the finite impulse response filter implementation and use this information to obtain greater power consumption savings.

Nevertheless the earlier embodiment described with reference to Figure 1, is likely to be preferable wherever it is necessary to provide a transmit signal for digital to analogue conversion which is sampled more frequently than twice per symbol (chip). In this case the architecture of Figure 1 can be used with the test filters 18, 20, sampled twice per symbol (chip) but the Pulse Shaping Filters 62, 64, can be oversampled by whatever factor is required.

Further modifications may be made to the arrangements of the earlier embodiment of the present invention, as hereinbefore described with reference to Figures 1, 2, and 3, without departing from the scope of the invention. In particular, the following modifications may be made.

1. The test filters 18, 20, need not be as long as the pulse shaping filters 62, 64. Some significant truncation in the length of the test filters 18, 20, will only result in the clipping threshold being extended slightly.

2. Look up tables in the form of Read Only Memory (ROM) can be used to considerable advantage in a number of different ways: a) For transmitting a single CDMA code at constant amplitude a lookup table with an address length equal to the number of signal samples over the length of the test filters 18, 20, multiplied by 2 (one address length for I signal component, one for Q) can be used. This can generate either values for c,, and c,, or, more advantageously, the clipped and filtered signal samples.

b) For transmitting a single code at constant amplitude with a known fixed direct sequence spreading code, the known code may be directed into sub sections of length equal to the test filter length. Thereafter a look up table may be generated for c,i and CQ, over a relevant length which thereby allows a smaller lookup table to be used. To allow for the effects of a modulation scheme two look up tables may be provided (e.g. or BPSK modulation of a QPSK code), one corresponding to modulation with a string of ones, the other corresponding to modulation with a string of reversals.

e) For a small number of codes (probably no more than two) of fixed amplitudes (not necessarily equal) a lookup table may be generated over all possible chip combinations, over the length of a shortened filter.

f) For a small number of codes of fixed amplitudes (not necessarily equal) a lookup table address may be provided for a filtered result over a time span, excluding the centre portion and the associated samples in the centre portion. For example, four bits I, four bits Q. three bits I & Qfor both centre samples would therefore require sixteen bits for three signal case. This makes practical a lookup table subtraction for each value. In fact, a result of the pre-clipping operation, is that the values p0 and p1 will depend on the corresponding data sets on both the I and the Qchannels. Nevertheless, modest size lookup tables could cover the case of addition of several signals. The largest lookup table (the third) would need 24Nwords where N is the number of signals added together. As before, the signals added together need not have equal amplitude provided their relative amplitudes are fixed. For example, a sixteen bit address (64K word) will serve for a lookup table for the addition of four signals. For the addition of three signals, a twelve bit address would suffice - an, additional four bits could cover a range of signal amplitudes (with the three signals retaining fixed mutual relationships).

This could be extended to s > l, however the lookup tables could become impractically large.

Whenever more than one envelope peak of a plurality of signals, exceeds the clipping threshold within the span of the pulse shaping filter, the clipping process may produce a sub optimum result. The reason for this is that the overclipping computations are treated independently for every signal sample reduced in amplitude by the clipping function, hereinafter referred to as a clipping event. The approach detailed above tends to lead to excessive clipping in the case of multiple events, since the signal reduction due to the first event is not accounted for when computing the required signal reduction to handle the next event A variety of recursive and/or backtracking approaches can be envisaged which would cope with this situation. These require batch processing based on overlapped burst signal samples. One possible approach is detailed below, the principle of which involves the following steps: 1. Generate a batch of source complex impulse samples, Rs.

2. Apply the pre-clipping algorithm to these samples (as per Figure 1) to form the vector, Bs of signal samples designated.

3. Apply the batch of complex impulse samples to a two times oversampling filter (this could be done using a fast convolution via the Fourier Transform) to form the filtered vector, B, of signal samples designated.

4. Perform the following iterative steps over the samples in Bf 4.1 For the current sample determine whether its envelope exceeds the clipping threshold. If not, go to step 5.

4.2 Compute the real and imaginary excess data values as described earlier.

4.3 By addressing the relevant entries in B, compute the excess coefficients gl,i and gQJ 4.4 Scale the relevant entries in B5 by the excess coefficients g,, and g,i to produce the signals for subtraction.

4.5 Take the signals for subtraction and multiply them by each of the filter coefficients and subtract them from the appropriate positions in BJ.

5. If last sample in B1 processed, step 6, otherwise go to step 4.

6. Output B1 overlapped with the Bf generated on the previous batch iteration and go to step 1.

This approach provides a facility whereby every overclipping operation takes full account of the effect of the previous overclipping operations. Nevertheless, this approach may be slightly sub-optimum, for example, in the case where a small excess data is removed, before removal of a large excess data in an adjacent or close position. A more powerful (albeit rather complex) process would result by modifying the above process as follows: Steps 1 t o 3 as before 4. Set the current sample to be the sample with the highest envelope ins.

4.1 For the current sample determine whether its envelope exceeds the clipping threshold. If not, go to step 6.

4.2 Compute the real and imaginary components of the excess data values as described earlier.

4.3 By addressing the relevant entries in By compute the excess coefficients g,, and gQ,i 4.4 Scale the relevant entries in Bs by the excess coefficients g,i and gQi to produce the signals for subtraction.

4.5 Take the signals for subtraction and multiply them by each of the filter coefficients and subtract them from the appropriate positions in B.

5. Go to step 4 6. Output B, overlapped with the B, generated on the previous batch iteration and then go to step 1.

The above process operates to process the signal samples working down through the highest peaks. It is likely that some of the lower peaks adjacent to the higher peaks will be reduced from above to below the clipping threshold merely by the operation of removing the higher peaks.

The above approaches may not be as unattractive as might first appear. The computing power required for this type of batch processing can be estimated on the basis of for example the 99 percentile of the number of envelope levels which exceed the clipping threshold over the batch interval. Any which cannot be processed in this time will be left unaltered. In practice, this may not matter, particularly if the ordered algorithm is applied since the remaining peaks will be relatively small. This would allow the batch processing duration to be maintained to a constant time.

Since the number of peaks exceeding the threshold will, in most circumstances be significantly fewer than those below it, there could be some benefit over the architecture of Figure 1, which must be designed to have enough processing power to cover the case where every sample exceeds the clipping threshold.

The embodiment of the present invention as hereinbefore described above can readily be extended to offset QPSK operation.

In this case, two times oversampling is natural with the on symbol (chip) positions corresponding to a particular clock cycle on the real channel at the same time as the off symbol (chip) position applies on the imaginary channel and vice versa. In this case, therefore, every clock cycle (2 per symbol (chip)) is treated independently. Whenever the envelope exceeds the peak, the excess data in the real and imaginary channels are computed as before. If the real channel corresponds to an on symbol (chip) then the modified coefficients are computed as per the first half of step 4 and the imaginary channel (which must correspond to an off symbol (chip)) will be computed as per the second half of step 4 and vice versa if the real channel corresponds to an off symbol (chip).

The affect of reducing the peak-to-mean amplitude ratio of a signal in accordance with the present invention may have an effect of producing very small degradations in Eb/No performance, of the received signal even with quite severe clipping.

Nevertheless, it may be desirable to reduce or, in some cases, to remove, even this small degradation. This can be achieved in the case of Direct Sequence Spread Spectrum modulation by use of a receiver provided with a matched filter to the clipped signal. The clipping function can be re-produced in a receiver by analogy to the processes operated in the transmitter. Consider the case of transmitting a single signal. Most of the circuitry in Figure 1 can be reproduced in the receiver, providing a means for reproducing the clipping function. The 1 & Q signal samples are generated by local (synchronised) code generators in the receiver, with the outputs of the fourth multiplier pair 44, 46, (stage 0) being fed, not into the pulse shaping filters 60, 62 (which is not required) but rather to correlators (with a suitable matching delay in the receive path). This implements an approximate matched filter to the transmitted signal. The reason it is only approximate is that the clipping function in the transmitter bases its clipping computations on the modulated code sequence whereas the receiver can only base its clipping computation on the unmodulated code sequence. Thus it is possible that some of the chips close to the symbol boundary will be clipped differently in the transmitter than in the receiver if there is a symbol reversal (either in lor Qor both). This effect is unlikely to be significant but can in any case be removed by generating two despreading waveforms (by reproducing the circuitry shown in Figure 1), one with an unmodulated code sequence, the other modulated with a series of phase reversals. The signal is then despread in parallel using both despreading signals. In fact, for a given symbol, inordinate signal samples may arise at the start and the end of the symbol, so there are four possibilities which must be handled.

This may be done by relying on the fact that, the clipping function for all cases will be the same at the centre of a symbol. For the possible clipping operation which occurs at the start of a symbol, a corresponding signal sample may be produced by switching over to the other waveform generator (and not) halfway through each correlation period. This gives a total of four possible despread symbols. A simplified Viterbi algorithm may be used (as familiar to those versed in the art) to compute the maximum likelihood sequence from these sequences of four symbols.

In principle, this procedure could be extended to cover the clipping of the sum of a number of signals. In this case code generators would be needed for all mutual relationships between bit sequences as well as all reversal conditions. Some simplification could be obtained by considering only mutual relationships and ignoring the reversal effects. This approach is unlikely to be practical beyond two signals.

For codes of modest length all pre-clipped code sequences could be pre-computed and stored in lookup tables. This would be practical provided there was not a large family of codes or if the family had simple mutual relationships (e.g. each code being a simple rotation of another code).

Claims

CLAIMS:

1. A method of reducing a peak to mean amplitude ratio of a signal represented as a plurality of discrete time signal samples at an output of a pulse shaping filter hereinafter referred to as the post-filtered signal samples, the said method comprising the steps of; (i) generating test signal data samples in dependence upon a plurality of discrete time signal samples representative of the signal before passing through the pulse shaping filter hereinafter referred to as the pre-filtered signal samples, the post-filtered signal samples being representative of a result of convolving the pre-filtered signal samples with an impulse response of the pulse shaping filter, and (ii) scaling the pre-filtered signal samples in accordance with the test data samples, so that the post-filtered signal samples are substantially constrained not to exceed to a predetermined threshold.

2. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claim 1, wherein step (i) comprises the step of; (iii) generating test signal data samples by convolving the prefiltered signal with an impulse response appertaining to the impulse response of the pulse shaping filter, and wherein the said method further includes between steps (i) and (ii) the steps of; (iv) identifying which of the test signal data samples have a modulus which exceeds a predetermined threshold hereinafter referred to as inordinate samples; (v) generating a plurality of modifier coefficients in accordance with the inordinate samples and the said pre-filtered signal samples, which modifier coefficients when used to scale the said pre-filtered samples substantially reduces the amplitude of the inordinate samples to the pre-determined threshold; and wherein step (ii) comprises the step of; (vi) arranging for the pre-filtered signal samples to be scaled with the modifier coefficients before being convolved with the impulse response of the pulse shaping filter, thereby reducing the peak to mean ratio of the said post-filtered signal.

3. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claim 2, wherein the pre-filtered signal samples are representative of a plurality of data symbols, and wherein the impulse response of the pulse shaping filter represents a duration corresponding to a plurality of symbols.

4. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claim 3, wherein each symbol period of the impulse response is represented by a plurality of impulse response coefficients.

5. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claim 4, wherein step (v) of the said method further comprises the steps of; (vii) calculating the modulus of each of a plurality of test signal data samples representative of a symbol, and assigning a complex sample corresponding to the test signal data sample with the greatest modulus to a complex variable L = L, + jLQ, (viii) identifying whether the test signal data sample represented by the complex variable L is an inordinate signal sample, by comparing the modulus of L with a threshold k, and if the modulus of L is less than or equal to k, setting all modifier coefficients to unity, and if not, proceeding with steps (ix) to (xi), the signal sample corresponding to L being an inordinate signal sample, (ix) calculating complex excess data given by E, = L,(1 k/lLI) and EQ = LQ(l - k/ILl) where ILI is the modulus of L, and representative of an amount by which the corresponding real and imaginary components of the inordinate signal sample L, must be reduced so that the modulus of L is reduced to that of the threshold k, (x) calculating the modifier coefficients in accordance with the pre-filtered signal samples, the impulse response of the pulse shaping filter, and the excess data, so that when the modifier coefficients scale the pre-filtered signal samples, the inordinate signal sample L is reduced by an amount corresponding to the excess data.

6. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claim 5, wherein step (x) of the said method further comprises the steps of; (xi) comparing the polarity of each of the real and imaginary components of a plurality of contribution data samples to the polarity of the corresponding real and imaginary components of the excess data, each of which plurality of contribution data samples corresponds to one of the plurality of pre-filtered signal samples scaled by a corresponding impulse response coefficient of the impulse response of the pulse shaping filter, and (xii) if the polarity of the real component of the contribution data sample is opposite to the real component of the excess data, setting the corresponding value of the real modifier coefficient to unity, and (xiii) if the polarity of the real component of the contribution data sample is the same as the polarity of the real component of the excess data, calculating the corresponding real modifier coefficient in accordance with a combination of the corresponding real contribution data sample scaled by the sum of all real contribution data samples squared and the real component of the excess data El, and (xiv) if the polarity of the imaginary component of the contribution data sample is opposite to the imaginary component of the excess data, setting the corresponding value of the imaginary modifier coefficient to unity, and, (xv) if the polarity of an imaginary component of the contribution data sample is the same as the polarity of the imaginary component of the excess data, calculating the corresponding imaginary modifier coefficient in accordance with a combination of the corresponding imaginary contribution data sample scaled by the modulus of the sum of all imaginary contribution data samples squared and the imaginary component of the excess data Ea,

7. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claim 6, wherein the plurality of impulse response coefficients representative of the impulse response of the pulse shaping filter are representative of a duration corresponding to a plurality of symbols, which symbols are comprised of two coefficients per symbol, and wherein the test signal data samples representative of the output of the test filter correspond to a single symbol, which single symbol is represented by two signal samples.

8. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claim 7, wherein step (x) of the said method further comprises the steps of; (xvi) identifying whether the inordinate signal sample corresponds to an on-symbol position, hereinafter referred to as an even sample, or whether the inordinate signal sample corresponds to an off-symbol sample position hereinafter referred to as an odd sample, (xvii) calculating the modifier coefficients corresponding to steps (x) to (xv), in accordance with a first process if the inordinate signal sample is an even sample, or calculating the modifier coefficients corresponding to steps (x) to (xv), in accordance with a second process if the inordinate sample is an odd sample.

9. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claim 8, wherein the first process operates to scale an odd number of pre-filtered signal samples (2s-1) with corresponding modifier coefficients numbered from-s +1 to s - 1 inclusive, in accordance with steps (x) to (xv) of the said method, and wherein the modifier coefficients are calculated in accordance with an amount by which all (2s 1) pre-filtered signal samples must be reduced, in order to respectively reduce the real and imaginary components of the inordinate signal sample by the real and imaginary excess data components E1 and EQ, where s is any integer.

10. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claim 8 or 9, wherein the second process operates to scale an even number of pre-filtered signal samples (2s) with corresponding modifier coefficients numbered from-s + I to s inclusive, in accordance with steps (x) to (xv) of the said method, and wherein the modifier coefficients are calculated in accordance with an amount by which all cos ) pre-filtered signal samples must be reduced, in order to respectively reduce the real and imaginary components of the inordinate signal sample by the real and imaginary excess data components E1 and EQ.

11. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claim 8, 9 orlO, wherein step (xiii) of the first process comprises the steps of; calculating each of the real modifier coefficients for the i-th prefiltered signal sample in accordance with the following equation;

and step (xv) comprises the step of calculating each of the imaginary modifier coefficients for the i-th pre-filtered signal sample in accordance with the following equation;

and step (xv) further comprises the step of assigning the modifier coefficients indexed, s, to unity, thereby causing them to be unmodified and so that cos =1 and CQ,5 =1.

12. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claim 9, 10 orll, wherein step (xiii) of the second process comprises the steps of; calculating each of the real modifier coefficients for the i-th prefiltered signal sample in accordance with the following equation;

and wherein step (xv) of the second process comprises the step of calculating each of the imaginary modifier coefficients for the i-th pre-filtered signal sample in accordance with the following equation;

13. A method of reducing a peak to mean amplitude ratio of a signal as claimed in any preceding Claim, wherein the modifier coefficients for scaling the real component of the pre-filtered signal sample are substantially equal to the modifier coefficients for scaling the imaginary component of the pre-filtered signal sample.

14. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claims 8 to 13, wherein step (xvi) is limited to identifying whether the even sample is an inordinate signal sample and wherein step (xvii) is limited to calculating modifier coefficients in accordance with the first process, when the even sample is an inordinate signal sample.

15. A method of reducing a peak to mean amplitude ratio of a signal as claimed in Claims 8 to 13, wherein step (xvi) is limited to identifying whether the odd sample is an inordinate signal sample and wherein step (xvii) is limited to calculating modifier coefficients in accordance with the second process, when the odd sample is an inordinate signal sample.

16. A method of reducing a peak to mean amplitude ratio of a signal as claimed in any preceding Claim, wherein the pre-filtered signal samples are representative of a non constant envelope signal, the said method further including the step of (xviii) preclipping the signal samples to the said predetermined threshold so that the amplitude of the pre-filtered signal samples is less than or substantially equal to the threshold.

17. Apparatus for reducing a peak to mean amplitude ratio of a signal represented as a plurality of discrete time signal samples at an output of a pulse shaping filter hereinafter referred to as the post-filtered signal samples, comprising a test filter means which operates to convolve pre-filtered signal samples representative of the signal before passing through the pulse shaping filter with an impulse response representative of the impulse response of the pulse shaping filter, a data processor being connected to the test filter means, which operates to generate a plurality of modifier coefficients, and a plurality of modification registers being connected to the data processor which operate to scale the pre-filtered signal samples with the said modifier coefficients, before being fed to the pulse shaping filter, thereby reducing the peak to mean ratio of the post filtered signal samples.

18. Apparatus for reducing a peak to mean amplitude ratio of a signal as claimed in Claim 17, wherein the modification registers comprise a plurality of scaling means for scaling the pre-filtered signal samples by the modifier coefficients connected thereto, and a plurality of delay shift registers which are connected to the said scaling means and arranged to provide discrete time displacement between the pre-filtered signal samples and the corresponding modifier coefficients.

19. Apparatus for reducing a peak to mean amplitude ratio of a signal as claimed in Claim 18, wherein the plurality of scaling means comprise a plurality of pairs of scaling means, a first scaling means of which pairs operates to scale real components of the pre-filtered signal samples by a modifier coefficient communicated thereto, and a second scaling means of which pairs operates to scale imaginary components of the pre-filtered signal samples.

20. Apparatus for reducing a peak to mean amplitude ratio of a signal as claimed in Claims 18 or 19, wherein the plurality of delay shift registers comprise a plurality of pairs of delay shift registers, which are connected and arranged so that a first delay shift register operates to delay real components of the pre-filtered signal samples by a clock cycle, and a second delay shift register operates to delay imaginary components of the pre-filtered signal samples by a clock cycle.

21. Apparatus for reducing a peak to mean amplitude ratio of a signal as claimed in Claim 20, wherein each pair of delay shift registers is interposed between pairs of scaling means, thereby providing a means for scaling the pre-filtered signal samples with the modifier coefficients wherein relative discrete time displacement between the said signal samples and corresponding inodifier coefficients is preserved.

22. Apparatus for reducing a peak to mean amplitude ratio of a signal as claimed in Claims 20 or 21, wherein one of the plurality of pairs of delay shift registers introduces a delay corresponding to a delay introduced by the test filters and the data processor into the pre-filtered signal samples.

23. Apparatus for reducing a peak to mean amplitude ratio of a signal as claimed in any preceding claim, wherein the test filter means comprises first and second test filters, a first of which test filters operates to convolve real components of the pre-filtered signal sample with an impulse response representative of the impulse response of the pulse shaping filter, and a second of which test filters operates to convolve imaginary components of the pre-filtered signal samples with an impulse response representative of the impulse response of the pulse shaping filter.

24. Apparatus for reducing a peak to mean amplitude ratio of a signal as claimed in Claim 23, wherein a final pair of the plurality of pairs of scaling means communicate the pre-filtered signal samples after scaling by the modifier coefficients to the pulse shaping filters.

25. Apparatus for reducing a peak to mean amplitude ratio of a signal represented as a plurality of discrete time signal samples at an output of a pulse shaping filter hereinafter referred to as the post-filtered signal samples, comprising a test filter means which operates to generate test data samples representative of a result of convolving pre-filtered signal samples representative of the signal before passing through the pulse shaping filter with an impulse response representative of the impulse response of the pulse shaping filter, a data processor being connected to the test filter means, which operates to generate a plurality of excess coefficients, and a plurality of modification registers being connected to the data processor and to the test filter means which operates to scale the test data samples with the said excess coefficients, thereby reducing the peak to mean ratio of the post filtered signal samples.

26. Apparatus as claimed in Claim 25, wherein the modification registers comprise a plurality of scaling means for scaling the prefiltered signal samples by the excess coefficients connected thereto, and a plurality of delay shift registers which are connected to the said scaling means and arranged to preserve discrete time displacement between the pre-filtered signal samples and the corresponding excess coefficients.

27. Apparatus as claimed in Claim 26, further comprising excess filter means being connected to a final pair of scaling means of the modification registers and a subtracting amplifier means, a negative input of which subtracting amplifier means being connected to an output of the excess filter means, and a positive input of the subtracting amplifier being connected to an output of the test filter means.

28. A transmitter comprising apparatus as hereinbefore described for reducing a peak to mean amplitude ratio of a signal to be transmitted by the transmitter.

29. A transmitter comprising a look up table and an addressing means, wherein the look up table comprises a plurality of postfiltered signal samples, which post filtered signal samples are pregenerated in accordance with the method for reducing a peak to mean amplitude ratio of a signal as hereinbefore described and in accordance with a pre-determined plurality of pre-filtered signal samples, and wherein the addressing means operates to generate a reference point in the look up table whereat a set of postfiltered signal samples are stored, in accordance with a corresponding set of pre-filtered signal samples.

30. A transmitter as claimed in Claim 29 wherein the look up table is a read only memory.

31. A receiver for detecting and recovering a transmitted spread spectrum signal which has been clipped in accordance with the method of reducing the peak to mean amplitude ratio of a signal as hereinbefore described, comprising a reference code generator which operates to generate a signal representative of a signal used to spread the spectrum of the transmitted signal, a signal processor which operates to generate a modified pre-filtered signal in accordance with the method of reducing a peak to mean amplitude ratio of a signal as hereinbefore described, a radio frequency down conversion unit which operates to detect and down convert the transmitted signal to a baseband signal representative of the transmitted signal substantially at baseband frequencies and a plurality of correlators, each of which plurality of correlators are connected to the radio frequency down conversion unit and which correlators operates to correlate the baseband signal with a phase shifted version of the modified prefiltered signal, which phase shifted version corresponds to a possible phase shift introduced by a modulating signal, and a second data processor connected to each of the said correlators which second data processor operates to perform a maximum likelihood sequence estimation process, thereby providing a means for detecting the most likely sequence of modulating symbols corresponding to the transmitted signal.

32. Apparatus for reducing a peak to mean amplitude ratio of a signal as hereinbefore described with reference to the accompanying drawings.