In maximum-or near-maximum-likelihood detection processes of the type frequently considered for use in applications where the digital data signal is received in the presence of both additive noise and severe intersymbol interference, the detector computes, for each of a number of possible sequences of received data symbols, the Euclidean or unitary distance between the corresponding received sequence, for the given signal distortion but in the absence of noise, and the sequence actually received. Such computations involve numerous operations of squaring or multiplication. The paper studies a general class of piecewise linear distance measures that involve no operations of squaring or multiplication, other than multiplication by the integer 2, leading to a useful simplification in the implementation of the detection process. The departure of each distance measure from the ideal is analysed theoretically for the simplest case of just one complex-valued received sample, and the results obtained are then extended to the general case of a message involving several received samples. Finally, results are presented of computer simulation tests assessing the effect of each distance measure on the tolerance of the detector to additive white Gaussian noise.
List of principal symbolsreal part of complex number imaginary part of complex number magnitude of complex number according to the specified distance measure magnitude of r,-p f according to the specified distance measure A, B, C, D or E magnitude of r { -p t according to the specified distance measure 1, 2, 3 or 4 distance between the vectors R and P according to the specified distance measure number of components in Y / 17 number of levels of data-symbol s,-number of components in R possible value of q t -component row vector whose/th component is p / value of resultant (total) data-signal component in r t /^-component row vector whose /th component is g,-a possible (but erroneous) vector Q /th received samplê -component row vector whose /th component is r { /th data-symbol (n -^-component row vector whose /th component is s t the possible (but erroneous) vector S corresponding to the vector Q e possible value of w,--component row vector whose /th component is w, -Gaussian noise component in r,-n-component row vector whose /th component is w,-possible value of 5, -(n -^-component row vector whose /th component isx,-(/ + l)th component of Y sampled impulse response of channel variance of w,-