Skip to main content

A Normal Limit Theorem for Power Sums of Independent Random Variables

01 November 1967

New Image

In many areas of transmission engineering, logarithms of sums of powers are considered in the form 1 = 10 Iog10 [10- W10 + · · · + 10 W 1 ° ] , where X1, . . ., X,, are random variables. Specifically, if A'i, . . are power levels in dB such that X,- = 10 log10 (w,/w u ) j = 1,2, ,n, Ar,, where w0 , wx , · · · , w,, are powers (e.g., expressed in watts), then the power level in dB of the sum w = wl + · · * + w,, is given by the socalled "power sum," Pn = 10 log10 (w/w0) = 10 log10 [10 Y,/1 ° + · · · + 10- W10 ]. Quite often X^, · · · , X,, arc taken to be mutually independent random variables with specified distributions, and it is of interest to determine properties of their power sum P,,. A major difficulty encountered in working with power sums is that the distribution and moments of such a sum usually cannot be ex2081 2084 T H E BELL SYSTEM T E C H N I C A L J O U R N A L , NOVEMBER 19(57 pressed in simple closed form. This includes, for example, the important case when X, · · · , Xn are mutually independent and each has a truncated normal distribution. Even in the simpler case when Xlf · · · , Ar,, are mutually independent, identically distributed, and Xx is normal, the problem is intractable. The difficulty and importance of the general problem, in turn, has led to a number of methods for approximating the distribution of a power sum.1-2> 3*4> 0-0i 8> 9 In the present paper, the asymptotic distribution of a power sum is studied. The main result is a limit theorem which shows that under very general conditions on the components Xi, X2, · · · , the corresponding power sums P,, will be asymptotically normal as n --» oo.