Abstract
Summary form only given, as follows. The authors describe a new algorithm for designing switched scalar quantizers for hidden Markov sources. The design problem is cast as a nonlinear optimization problem. The optimization variables are the thresholds and reproduction levels for each quantizer, and the parameters defining the next-quantizer map. The cost function is the average distortion incurred by the system. The next-quantizer map is treated as a stochastic map so that all of the optimization variables are continuous-valued, allowing one to use a gradient-based descent procedure. Details are given for computing the cost function and the required gradients. Numerical simulation results are presented that compare the new system to standard adaptive quantizers for both long-term SNR and segmental SNR. It is observed that the optimal system usually requires a stochastic next-state map and sometimes requires nonsymmetric quantizers even though the subsource densities are symmetric.
Original language | English |
---|---|
Pages | 138 |
Number of pages | 1 |
State | Published - 1988 |