Abstract: A method and a device for training the equalizer of a receiver used in a QAM transmission system, by means of an unknown data signal, thereby enabling in particular said equalizer to be trained in the cases described above. In accordance with the method of the present invention, in a data receiver wherein, under normal operating conditions, the data is detected with respect to a first constellation, the equalizer is trained by detecting the data with respect to a second constellation which comprises fewer points than said first constellation and by adjusting the coefficients of the equalizer by means of an error signal produced as a result of the detection operation performed with respect to the second constellation.In accordance with a first embodiment of the present invention, the second constellation is comprised of those points of the first constellation that are farthest from its center.

Abstract: A specialized processor capable of computing a sum of products S=.SIGMA..+-.Pi where every product Pi is the product of two n-bit complex operands Ai+j Bi, the multiplier, and Ci+j Di, the multiplicand, where j=.sqroot.-1. The processor includes an instruction storage, means for decoding instructions read out of said storage and for controlling the operation of the processor, a data storage, and a multiplication and accumulation unit which has two multiplier-accumulator devices and several buffers for storing the operands Ai, Bi, Ci and Di sequentially read out of data storage. The real part Ai and the imaginary part Bi of the multiplier are respectively applied to the Multiplier inputs of the multiplier-accumulator devices and the real part Ci of the multiplicand is applied to the Multiplicand inputs of the multiplier-accumulator devices, which simultaneously compute the products Ai Ci and Bi Ci.