An adaptive Decision Feedback Equalizer (DFE) (40) for a digital cellular mobile radio channel demodulator (30) employs a Complex Fast-Kalman Adaptation algorithm (56) to track channel variations. Sensitivity to sample timing jitter is reduced by providing the DFE with fractionally spaced feed-forward taps (50'). Deficiencies inherent in using a reduced precision implementation are overcome by adding a dither signal to sets of operations in the algorithm, the dither signal comprising an appropriately selected Gaussian random variable. For small delay spreads of approximately one third of a symbol duration or less, a resulting degradation in Bit Error Rate is avoided by switching the DFE out of the circuit or by reducing the number of taps of the DFE. For delay spreads of less than 40 microseconds and greater than approximately 10 microseconds, a (2,3) fractionally spaced DFE provides an adequate compromise between complexity and performance.

The object of the present invention is to provide an automatic equalizer which can automatically equalize distortion including a precursor component with a small amount of calculation and a small circuit scale. The automatic equalizer includes a subtractor for calculating differences between a received signal and N estimated received signals and outputting N estimated error signals, a decision unit for outputting part of a transmission signal sequence corresponding to one of the N estimated error signals which exhibits the lowest absolute value as a detected signal as the output of the automatic equalizer, a tap coefficient generation circuit for receiving the received signal as an input thereto to produce an impulse response and outputting the impulse response as a first tap coefficient group corresponding to a postcursor component of the impulse response and a second tap coefficient group corresponding to a precursor signal of the impulse response, a transversal filter for receiving the detected signal and the first tap coefficient group as inputs thereto and outputting a postcursor estimation signal, and an estimated received signal generator for receiving the postcursor estimation signal and the second tap coefficient group as inputs thereto and outputting the N estimated received signals corresponding to possible N transmission signal sequences.

A method for use in a digital data transfer apparatus, such as a radio telephone (10), that includes an adaptive equalizer, such as an adaptive filter (32), for receiving, correcting, and outputting data symbols. The method determines correction coefficients for the adaptive equalizer and includes a first step of, for every data symbol output by the adaptive equalizer, maintaining a first matrix R that is a NxN channel correlation matrix and a second matrix P that is a N.times.1 cross correlation vector between a received signal and a transmitted symbol. The method includes a second step of, for every k.sup.th symbol output by the adaptive equalizer, determining a set of correction coefficients (C) for the adaptive equalizer based upon the matrices R and P. A third step of the method applies the set of determined correction coefficients to the adaptive equalizer so as to optimally correct a subsequently received data symbol. The correction coefficients are calculated with an inverse matrix of the channel correlation matrix. As a result, the correction coefficients are, in principle, optimal.

Disclosed is a hardware arrangement for equalizing waveform distortion caused by multipath fading in a digital radio communications system, which hardware arrangement uses a recursive least-squares algorithm into which a forgetting factor is introduced. An incoming signal (PSK signal for example) is applied to an automatic gain controller which generates a first signal whose value varies with a signal-to-noise ratio of the incoming signal. A forgetting factor controller is coupled to the automatic gain controller to receive the first signal and generates the forgetting factor the value of which varies with the value of the first signal. An adaptive equalizer receives the incoming signal and the forgetting factor. The adaptive equalizer equalizes the waveform distortion using the forgetting factor according to the recursive least-squares algorithm.

An automatic equalizer that can realize high-quality signal equalization without erroneous decision in the environment where a sequence candidate tends to be erroneously selected due to noises because the component corresponding to a demodulation point is close to the precursor component in magnitude. The automatic equalizer outputs the absolute values of impulse response signals, determines as candidate signals response absolute value signals larger than a threshold value, selects as a demodulation point a point corresponding to a candidate signal with the smallest propagation value among the candidate signals, and decides the selected point as a demodulation point setting signal.