Showing papers on "Concatenation published in 1973"

Coupled Decoding of Block-Convolutional Concatenated Codes

Abstract: Straightforward implementation of a maximum likelihood decoder implies a complexity that grows algebraically with the inverse of error probability. Forney has suggested an approach, concatenation, for which error probability decreases exponentially with increasing complexity. This paper presents the results of an evaluation of a particular concatenation system, structurally similar to the hybrid system of Falconer, employing a Reed-Solomon outer code and an inner convolutional code. The inner decoder is a Viterbi decoder of constraint length less than the corresponding encoding constraint length (nonmaximum likelihood). The outer decoder assumes one of three possible forms, all employing the likelihood information developed by the inner decoder to assist in outer decoding. Error corrections and erasure fill-ins achieved by the outer decoder are fed back to the inner decoder. Performance is evaluated through computer simulation. The three outer decoders are found to provide approximately the same performance, all yielding low error probabilities at rates somewhat above R comp of sequential decoding and at signal energy to noise density ratios per information bit around 1.7 dB.

The USC chess program

Abstract: Much attention has been given recently to the “linguistic” approach to pattern recognition. The basics ingredients of this approach are a set of primitive picture parts and a grammar whose rules compose the primitive parts into a class of pictures. The basic idea of linguistic pattern recognition is to generalize string grammars to two dimensions. This requires a generalization of “concatenation” to two dimensions. Several applications have resulted from the definition of simple picture grammars[1-4]. This field should properly be called structural pattern recognition since its basic goal is to study the processing of the structure of pictures. From this viewpoint, one may ask whether simple grammars and the present concatenation schemes are adequate for the processing of complex scenes, or whether other approaches should be sought. These questions may be answered in the affirmative by considering the case of chess. The structure of a chessboard appears to be incredibly complex, yet humans seem to recognize familiar situations by means of structural organization of the board. Thus, chess seems to be an ideal paradigm case for complex structural pattern recognition. The USC chess program was produced from these studies. The remainder of this report will give a brief description of the chess program itself.