文档介绍：84 IEEE TRANSACTIONS MUNICATIONS, VOL. COM-28, NO. 1, JANUARY 1980
An Algorithm for Vector Quantizer Design
YOSEPH LINDE, MEMBER. IEEE. ANDRES BUZO, MEMBER, EEE, AmROBERT , SENIOR MEMBER. EEE
In particular, the algorithm’s convergence properties are
’ Abstract-An efficient,and intuitive algorithm is presented for the design
of vector quantizers based either on a known prohabitistic model or on a demonstrated herein by several examples. We consider the
long training sequence of data. The basic properties of the algorithm are usual test cas. for suchalgorithms-quantizer, design for
discussed mid demonstrated by general distoriion memoryless Gaussian sources with a mean-squared. error distor-
measures andlong blocklengths are allowed, as exemplified by the designof
parameter vector quantizers of tendiensional vectors arising in Linear tion measure; but we design and evaluate block quantizers with
Predictive Coded (LE)pression plicated distortion a rate of one bit per symbol and with blocklengths,of1 through
measure arisiig in LPC analysis that does not depend only on the error 6. Comparison with recently developed lower bounds to the
vector. optimaldistortion of such block quantizers (which provide
strict improvement over the traditional bounds of rate-distor-
INTRODUCTION tion theory) indicate that the resulting quantizers are indeed
A-N efficient and intuitive algorithm for the design of good nearly optimal and not simply locally optimal. We also con-
block or vector quantizers with ‘quite general distortion sider a scalar case where local optima arise and show how a
measures is developed for use on either known probabilistic variation of the algorithm yields a global optimum. .
source descriptions or on a long training sequence of data. The The algorithm is also used to design a quantizer for 1Odi-
algorithm is based on in approach of Lloyd [I] ,is not a varia- mensional vectors- arising pressi