Direct-feedback coders: Design and performance with television signals

Direct-feedback coding is a refinement on the well-known differential coding method. Two filters are used at the transmitter of a direct-feedback coder; one connected in series with the input and the other in the forward path of a feedback loop that contains the quantizer. The first filter preemphasizes the signal and determines the overload characteristic of the coder; the other filter shapes the quantization noise and sets the stability of the feedback. At the receiver a filter reconstitutes the signal spectrum and deemphasizes the noise. For television the preemphasis should he a short time-constant differentiator, the deemphasis a short time integrator, and the feedback filter a long time integrator. Conventional differential coders use a single filter in the feedback path both to provide preemphasis and to shape the feedback characteristic, so the design is a compromise. Compared with direct-feedback coding they usually have less feedback gain and a larger time constant in the preemphasis and deemphasis, consequently, the contouring noise is more visible and the streaking caused by transmission error is longer. Although only application to television is considered, the methods have wider use. General formulae are given for the output noise and optimum filter characteristics; they take into account signal spectra, frequency weighting for noise, sampling rate, quantization step size, and an overload parameter. Measurements on real coders, operating on TV signals, and digital simulations confirm the results.