An iterative time-frequency domain algorithm for reduction of peak-to-rms ratio

In voice radio communication, reducing the peakto-rms ratio of a speech signal is of interest when there is a peak transmitter power limitation imposed by either economic or legal considerations. If the peak-to-rms ratio can be reduced, then average output power can be increased without increasing the peak transmitter power, resulting in “more bang for the buck,” so to speak. This is of particular interest during marginal radio communication conditions. In the past, this goal has mainly been achieved by analog signal processing; the technique usually involves either audio or radio frequency amplitude limiting. Of course, there is a trade-off between perceived loudness and perceived voice quality. In a recent paper [ 11, Quatieri and McAulay investigated this problem in great detail and proposed a digital signal processing peak-to-rms reduction algorithm based on a sinusoidal analysis/synthesis system. Their paper has served as a starting point and an inspiration for the work reported herein. This correspondence describes the initial work on an iterative time-frequency domain algorithm for reduction of peak-to-rms ratio. Time-frequency iteration has been used to solve other problems, such as signal reconstruction from phase [ 21. The basic idea is straightforward: in each domain there is a constraint which is imposed at each iteration. In the reconstruction from phase problem, the known phase is the frequency domain constraint, and the known time duration is the time domain constraint. In the peakto-rms reduction problem, the spectral magnitude of the given signal is the frequency domain constraint, and a constant envelope (defined below) is the time domain constraint. Initial experiments with the peak-to-rms reduction algorithm suggest that usable results may be obtained with only one or two iterations, and in some cases only one iteration may suffice. The spectral magnitude constraint is the last to be imposed, since the shape of the spectral magnitude is thought to be extremely important for intelligibility.