Abstract: A signal processor which can be configured to perform interpolation or decimation of an analog input signal, using a combination of continuous-valued discrete-time sampling together with digital filter coefficients. In its preferred embodiment, the signal processor includes a signal sampling circuit configured to produce a continuous-valued discrete-time sequence of the most recent N input samples, with the sequence being stored in a circular buffer arrangement of sample-and-hold circuits. Further, in its preferred embodiment, the signal processor includes a filter coefficient shift-register that stores digital filter coefficients, with the shift register contents being cyclically shifted to maintain the correct time correspondence with the input samples held in the circular buffer of input samples.

Abstract: Disclosed are bidirectional equalizer arrangements in which the outputs of the forward and reverse equalizers are combined using low complexity processing to yield a single output sequence. In each disclosed arrangement of the invention, the low complexity processing utilizes the equalizer error outputs of the forward and reverse equalizers, which are then zero-delay lowpass filtered to provide relative reliability estimates. In one arrangement of the invention, these reliability estimates are used to select, on a symbol-by-symbol basis, either the forward or the reverse equalizer output. In another arrangement of the invention, a single linear lowpass filter is used to filter the difference of the equalizer error signals. In another arrangement of the invention, the relative reliability estimates are used to form a weighted average of the forward and reverse equalizer pre-decision signals.

Abstract: A metric sifting, i.e., sorting or selection, method which can efficiently provide a sorted set of survivor metrics during breadth-first reduced-search decoding of convolutional codes. The preferred embodiment efficiently implements the M algorithm by providing a sorted set of M survivor metrics using a linear-time (O(M)) number of comparisons. To obtain this efficiency, the method employs partitioning of branch metrics into implicitly sorted subsets, and employs efficient merging of these subsets. Compared to the prior art, the number of comparisons for metric sifting during M algorithm decoding of typical rate 1/n binary convolutional codes is reduced by 30-40%; more specifically, the number of comparisons is reduced from (2n-1)M-(n-1) to n(M-1)+2.sup.n -1.