This paper describes the design of a high quality 16 Kb/s speech processing scheme known as the Split-Band Adaptive Predictive Coding (SBAPC). In this algorithm, the frequency contents of the input signal are evenly split into two subbands via quadrature mirror filters. Then, adaptive predictive coding (APC) is applied to the subband waveforms. A bit allocation scheme that dynamically alters the quantizer bit assignments according to the energies of the two bands is employed in order to improve further the speech quality. Incorporating forward error correcting codes and noise suppression procedures, the quality of the SBAPC processed speech has been maintained for high background noise levels and channel error rates.

The performance of a noncoherent frequency-shift keyed (FSK) system is analyzed in the presence of multipath fading by using a sinusoidal series expansion technique for the representation of a bandlimited Gasussian process. The probability of error is calculated as a function of the system parameterBT, signal-to-noise ratio, fading bandwidth and direct to reflected signal power ratio.

A Gauss quadrature rule (GQR) is used to develop an efficient series representation of a band-limited Gaussian process. This representation allows one to analyze the performance of an incoherent frequency-shift keyed (FSK) system under the influence of an additive Gaussian noise. This system consists of two parallel filters followed by two squarers, with the sum of their outputs fed to an integrate-and-dump detector. Numerical results are presented for some cases of practical interest.

The effect of a hard limiter on a quadrature PSK (QPSK) signal is determined for a sample detector and an integrate and dump detector. In the latter case theBTproduct is assumed to be infinite. The degradation due to hard limiting is approximately 1 dB.