Showing papers by "Codex Corporation published in 1987"

Real-time vector APC speech coding at 4800 bps with adaptive postfiltering

Abstract: An improved Vector APC (VAPC) speech coder at 4800 bps produces speech with very good communications quality while maintaining a complexity low enough to allow a real-time implementation with at most two commercially available DSP chips. The VAPC algorithm combines APC with vector quantization and incorporates analysis-by-synthesis, perceptual noise weighting, and adaptive postfiltering. A novel adaptive postfiltering technique helps to achieve an essentially inaudible level of coding noise. Real-time software has been developed for an implementation using the AT&T DSP32 floating-point processor chip. The overall complexity of the implemented VAPC system is about 3 million multiply-adds/second of computation and 6 kwords of memory.

Gain-Adaptive Vector Quantization with Application to Speech Coding

Abstract: The generalization of gain adaptation to vector quantization (VQ) is explored in this paper and a comprehensive examination of alternative techniques is presented. We introduce a class of adaptive vector quantizers that can dynamically adjust the "gain" or amplitude scale of code vectors according to the input signal level. The encoder uses a gain estimator to determine a suitable normalization of each input vector prior to VQ encoding. The normalized vectors have reduced dynamic range and can then be more efficiently coded. At the receiver, the VQ decoder output is multiplied by the estimated gain. Both forward and backward adaptation are considered and several different gain estimators are compared and evaluated. Gain-adaptive VQ can be used alone for "vector PCM" coding (i.e., direct waveform VQ) or as a building block in other vector coding schemes. The design algorithm for generating the appropriate gain-normalized VQ codebook is introduced. When applied to speech coding, gain-adaptive VQ achieves significant performance improvement over fixed VQ with a negligible increase in complexity.

Polyphase filter matrix for rational sampling rate conversions

Abstract: The polyphase filter array has been used for efficient implementations of filters with integer sampling rate conversions. [1] The filter in the high sampling rate side is decomposed into its polyphase filters which can be moved to the lower sampling rate side without changing their functions. For FIR filters the computational complexity is reduced by a factor equal to the sampling rate ratio. A rational (L/M) sampling rate conversion system realized with a 1-to-L interpolator followed by an M-to-1 decimator has three sampling rates F, LF and (L/M)F involved. By using the polyphase filter array a filter operating at the sampling rate of LF can be implemented in either the input side or the output side with lower sampling rates. The polyphase filter matrix structure will operate at the sampling rate of F/M, which does not show in the above model and is lower than any one of those three rates. For FIR filters the computational complexity is reduced by a factor of LM compared to the direct realization of the integral filter or by a factor of M (or L) compared to the polyphase filter array realization while the system input-output relation is maintained.

Efficient time-recursive least-squares algorithms for finite-memory adaptive filtering

Abstract: This paper deals with efficient algorithms in the sense of minimization of the computational complexity for least-squares (LS) adaptive filters with finite memory. These filters obtain the current estimate of the desired response using only a fixed finite number of past data. First, two new fast recursive least-squares algorithms with computational complexities 14m and 15m multiplications and divisions per recursion (MADPR), respectively, are introduced ( m is the filter order). Then a new estimation-error-oriented recursive modified Gram-Schmidt (RMGS) scheme with a complexity of 2m^{2} + 10m MADPR is given. Finally, the learning characteristics of these algorithms are discussed and some simulation results are included.

Covariance and autocorrelation methods for vector linear prediction

Abstract: A novel least-squares formulation of the vector linear prediction (VLP) problem is presented. Based on this formulation, we develop two new design methods for obtaining the optimal vector predictor for frame-adaptive prediction: the covariance method and the autocorrelation method, which bear the names of the corresponding methods in scalar LPC analysis. Our formulation reveals several previously unrecognized properties of the resulting normal equation. Simulation results for VLP of speech waveforms confirm that the two proposed methods indeed give higher prediction gain than previously developed methods.

Decoding of lattices and codes

Abstract: Simplified decoding procedures based on trellis diagrams are given for lattices and codes. In one arrangement, for selecting a codeword near to a given N-tuple r, the codeword representing a point in an N-dimensional lattice, the N-tuple r comprising a sequence of N real values ri representing signals, the values r of the N-tuple r being organised into n sections r; respectively of lengths Nj, I ≦ j ≦ n, where n ≧ 3, and Nj < N, a decoder comprises at least first and second decoding stages. The first decoding stage comprises substages associated respectively with the sections rj, I ≦ j ≦n.Each such substage of the first stage comprises means for evaluating distances between multiple sets of possible partial codewords and the section rj, these partial codewords comprising the corresponding N of the symbols appearing in a subset of said codewords. Additionally, each first stage substage has means for selecting as a survivor from each of the multiple sets one of the partial codewords. The selection is based on the distances. Each substage also includes means for providing to the second decoding stage information indicative of each of the multiple survivors and its distance. The second decoding stage comprises n-I substages. Each such substage of the second stage corresponds to a supersection made up of a combination of two parts, each such part being a section or a supersection combined in an earlier substage of the second stage, a final such substage corresponding to a final supersection whose parts make up the complete N-tuple r. Each substage of the second stage except the final substage comprises means for evaluating distances between multiple sets of possible partial codewords and the received values in the corresponding supersection. This evaluation is based on the information indicative of survivor distances for the survivors corresponding to each of the two said parts. Each substage of the second stage other than the final substage also comprises means for selecting as a survivor from each of the multiple sets one partial codeword on the basis of the distances, and for providing information indicative of each of the multiple survivors and its distance to subsequent decoding substages of the second stage. The final substage comprises means for evaluating distances between said codewords and the N-tuple r. This evaluation is based on the indications of survivor distances for survivors corresponding to each of two of the parts. The final substage also includes means for selecting one codeword as a finally decoded codeword based on the distances between codewords and the N-tuple r, and for providing information indicative of the finally decided codeword as an output of the decoder.