Title :
On the cost of finite block length in quantizing unbounded memoryless sources
Author :
Linder, Tamás ; Zeger, Kenneth
Author_Institution :
Dept. of Math. & Comput. Sci., Tech. Univ. Budapest, Hungary
fDate :
3/1/1996 12:00:00 AM
Abstract :
The problem of fixed-rate block quantization of an unbounded real memoryless source is studied. It is proved that if the source has a finite sixth moment, then there exists a sequence of quantizers Qn of increasing dimension n and fixed rate R such that the mean squared distortion Δ(Qn) is bounded as Δ(Qn )⩽D(R)+O(√(log n/n)), where D(R) is the distortion-rate function of the source. Applications of this result include the evaluation of the distortion redundancy of fixed-rate universal quantizers, and the generalization to the non-Gaussian case of a result of Wyner on the transmission of a quantized Gaussian source over a memoryless channel
Keywords :
memoryless systems; quantisation (signal); rate distortion theory; redundancy; source coding; distortion redundancy; distortion-rate function; finite block length cost; finite sixth moment; fixed-rate block quantization; mean squared distortion; memoryless channel; nonGaussian case; quantized Gaussian source; quantizing unbounded memoryless sources; sequence of quantizers; source coding; universal quantizers; Convergence; Costs; Laplace equations; Memoryless systems; Performance analysis; Propagation losses; Quantization; Rate distortion theory; Redundancy; Source coding;
Journal_Title :
Information Theory, IEEE Transactions on