To compensate for linear distortions, e. g. two-path fading, in QAM (quadrature amplitude modulated) systems usually LMS (least mean squares) equalizers are used. In some radio systems, a modulation format with constant envelope is used, e. g. CPM (continuous phase modulation), in order to be able to use inexpensive RF components. With this modulation, the transmitter amplifiers can be driven to the saturation point.

For the transmission of signals there exist standards which require a certain acceptance of linear distortions. Therefore, in some transmission systems, equalization is necessary.

In CPM systems, there is usually no decision in the I/Q (inphase and quadrature component) domain, but a processing in the phase domain. This has the advantage of a direct processing of the relevant information (the phase) in CPM, but the disadvantage that there do not exist decided I. and Q symbols, so that a standard LMS equalizer as in the QAM case cannot be used.

The underlying task of the present invention is to create a method and an apparatus which allows the equalization of linearly distorted CPM signals. The said objectives are achieved by the features from claims 1 respectively 9.

Further developments of the invention are characterized in additional claims.

In accordance with the invention, an adaptive filter is used as an equalization filter of CPM signals. The adaptation of its filter coefficients is based on instantaneous errors which are determined in the output signal of the filter. Furthermore, in order to stabilize the coefficients, the respective adaptation value of the filter coefficients will be driven back to a predetermined value with a defined time constant.

CPM is a modulation type with constant envelope. Since in the present invention only this signal property, namely the constant envelope, is used, the CMA (constant modulus algorithm) equalizer in accordance with the invention works advantageously independently from clock and carrier of the signal. The inherent tendency to instabilities of the fractionally spaced (T/2) equalizer is reduced substantially by the usage of a tap-leakage. Furthermore, the equalizer in accordance with the invention compensates also for I/Q- distortions (differences in orthogonality and gain) so that these settings need not be tuned in the demodulator.

The invention is described more detailed in the following figures and examples. It is shown: Fig. 1 A principle schematic of the structure of a CPM system with the model of a two-path fading channel in order to explain an application of the invention; Fig. 2 Basic principle of an equalizer; Fig. 3 Diagram to illustrate the derivation of the instantaneous error used for the'adaptation of the filter coefficients ; Fig. 4 Illustrative example for an error detector; Fig. 5 Example for the structure of a correlator; Fig. 6 Example of one of the components of the correlator according to Fig. 5 ; Fig. 7 Example of a correlator cell to be used in the component according to Fig. 6; Fig. 8,9 Measured curves to illustrate the effectiveness of the invention.

In the embodiment shown in Fig. 1, digital data samples are fed to the input of a modulator 2 which supplies CPM-modulated output signals !' ; to a transmitter 4 in which the digital CPM signals are converted into analog signals and modulated onto a carrier'by an up-converter. This modulated RF signal s (t) is then fed to the transmit antenna 6 and sent to the receive antenna 8. The signal is propagated on a direct path 10 and indirectly on echo paths 12, of which only one is shown here representatively.

The received signals at the receive antenna 8 are fed to a receiver 14 which performs the down-conversion and A/D- conversion (analog-digital) of the signal. The receiver supplies output signals x (t) to a signal processing unit 16, which consists of an equalizer 18, a prefilter and phase detection unit 20 ; ; and a phase processing and decoding unit 22.

At the output of the unit 16, the regenerated input data is present as output data.

Fig. 2 illustrates the basic principle of an equalizer (18).

The signals x (k) which were distorted in the transmission channel are fed to a filter 24, which can for instance be implemented as an FIR filter with adjustable filter coefficients. Its output signals y are fed to an error detector 26 which detects the instantaneous errors e (k) in the output signal and supplies them to a correlation and adaptation unit 28 which adjuststhe coefficients of the FIR filter based on the errors e and some reference signal r (k), shown here to be the filter input signal. The unit 28 also makes sure that if there exist only small correlation contributions, the adapted filter coefficients return with a defined time constant to a predetermined value (tap leakage). This way, the signal equalization is performed based on the detected signal errors, hence in continuous adaptation to the transmission channel quality.

The cost function which is minimized by the CMA equalizer in accordance with the invention is: Here, the mean output amplitude (the modulus) is normalized to 1 without restricting the invention.

The cost function becomes minimum when the deviation of the squared output amplitude from 1 is minimum. It is zero, when all complex output samples y (k) have the amplitude 1.

The adaptation algorithm defines an instantaneous error e (k) as the difference of the filter output signal value from its projection onto the unit circle, as it is shown in Fig. 3. It is therefore: Unlike in QAM systems, the instantaneous error does not point to the correct direction, since only the radial component of the error can be determined. By the averaging done in the coefficient adaptation, however, the information about the direction (phase) of the error is obtained again, The complex error has to be split into an I and Q part. With this instantaneous error then the correlation can be done as in a complex valued LMS equalizer following the stochastic gradient descent, : CiII(k+1)=CiII(k)-µ#eI(n+i)#xI(n) CQl (k+1)=CiQI(k)-µ#eI(n+i)#xQ(n) <BR> <BR> <BR> CiIQ(k+1)=CiIQ(k)-µ#eQ(n+i)#xI(n)<BR> <BR> <BR> <BR> <BR> CiQQ(k+1)=CiQQ(k)-µ#eQ(n+i)#xQ(n) where c (k) i-th equalizer coefficient in the II-rail at time . instant k'' e (n) I-part of the instantaneous error at time instant n step width XI (n) Equalizer input signal at time instant n, channel I As reported in literature (J. R. Treichler, V. Wolff, C. R.

, _, Johnson Jr.,"Observed misconvergence in the constant modulus adaptive algorithm", Conference Record of the Twenty-Fifth Asilomar Conference on Signals, Systems and Computers (Cat.

No. 91CH3112-0) IEEE Comput. Soc. Press : Los Alamitos, 1991. p. 663-7 vol. 2), the CMA equalizer does not always converge to a solution which equalizes the input signal.

To solve this problem, in accordance with the invention a tap leakage is provided which causes a little deviation of the coefficients from their optimum values in the case of a distorting channel, but makes sure that with a flat channel, the coefficients do not diverge too much from their nominal position.

As already stated in the explanation of Fig. 3, the error e (k) is detected by projection of the filter output signal onto the unit circle and splitting in I and Q parts. Since the filter output signal is already given in I and Q, the error can be directly calculated in I and Q: To save word length, the multiplication can also be performed with the sign of yI respectively yQ.

The structure of the error detector which implements the equations above for the case of a normalization of the output amplitude to 1 is shown in Fig. 4. The input signals f (k) and yQ (k), respectively, are each fed to a squarer in the form of multipliers 31 and33, respectively. The squared signals are added in an adder 35. From this sum, a constant, here 1, is subtracted in another adder 37. The result is multiplied in two multipliers with the input signals yI (kJ and yQ (k), respectively, in order to form the instantaneous errors el (k) and eQ (k), respectively.

Let us now consider the adaptation of the coefficients. The coefficients are adapted by the correlation of a reference signal and an error signal. Fig. 5 and Fig. 6 show the upper level structure of the correlator, whereas Fig. 7 shows the details of a correlation cell.

According to Fig. 5, the I and Q samples of the reference signal input are fed to two delay units 30a and 30b in order to compensate for the basic delay of the filter 24. The output signals of unit 30a are supplied, as I reference values, to the first inputs of respective correlation units 38 and 40, whereas the output signals of unit 30b are supplied, as Q reference values, to the first inputs of respective correlation units 42 and 44. The second inputs of correlation units 38 and 42 have coupled thereto the I error signal, whereas the second inputs of correlation units 40 and 44 have coupled thereto the Q error signal. The structure of the correlation units follows Fig. 6.

For an N-tap equalizer they consist in the illustrated manner of N correlation cells 46 and N-1 delay elements 48. By correlating reference values and error values, the coefficients for the filter 24 are adapted.

Fig. 7 shows the structure of a correlation cell 46. The error signal e (k) and the reference signal r (k) are multiplied in a multiplier 50. The resulting product is then amplified or attenuated in some unit 52. By an adder 54, a tap-leakage value is subtracted from this signal. The tap-leakage value is supplied by an adaptation unit 56 which adjusts the tap-leakage value, depending on the coefficient, to an appropriate value.

The output signal of the adder 54 is then accumulated in a resettable accumulator 58 with saturation and supplied as coefficient to the filter 24.

1 As mentioned before, tap leakage makes the filter coefficients return to a predetermined value if there exist no or only small correlation contributions. It is achieved by a subtraction (adder 54) of a small part of the accumulator output value. If the leakage values lie e. g. in the order of 2-20, the accumulator word length can be still kept small when the tap leakage value is not subtracted in every cycle, but e. g. only every 16th cycle. In this case the accumulator can be shortened by 4 bits with respect to a solution with a subtraction every cycle.

It is desirable and necessary to achieve a stable behavior in the undistorted case, in which only the center tap should be different from zero and all other taps should have values close to zero, and also to obtain a sufficient equalization with a distorting channel. In order to achieve this, the tap leakage can be set differently for each tap.

The equalizer in accordance with the invention improves the system behavior mainly in two ways: * It reduces the effects of two-path fadings (the signature area is reduced) * It compensates for mistuning of the I/Q demodulator (90° phase, gain difference between I and Q) In Fig. 8, the improving effect of the equalizer to two-path fadings is shown in a signature measurement. Without equalizer the minimum signature depth is 10 dB, whereas with equalizer the minimum is increased to 18 dB.

The system parameters in this case are: fair = 38.913 MHz h (modulation index) = 0. 3 T (echo delay) = 6. 3 ns Fig. 9 shows the improving effect on demodulator mistuning: Errors in the 90° phase tuning of the I/Q demodulator or,' relative gain differences between I and Q normally cause a distortion of the received signal. A circle in the T/Q plane would be transformed to an ellipse. Without countermeasures, this results in a phase error leading to an increased bit error rate.

By implementing the CMA equalizer as a complex valued 4-rail equalizer, these distortions can be compensated for without any additional effort, by releasing one cross coefficient at the position 0 (CIQ (0) or CQI (0)).

The measurement results in Fig. 9 illustrate that the CMA equalizer can compensate a phase error of 15'without substantial degradation. Since there is a direct relation between the gain difference and the phase error (a phase error of 15° corresponds to approximately 2.3 dB gain difference), the same holds accordingly for the compensation of I/Q gain differences.

Many variations and modifications may be made to the presented embodiment of the invention without departing substantially from the spirit and principles of the invention. All such modifications and variations are intended to be included herein within the scope of the present invention, as defined by the following claims.