MUSIC (multiple signal classification) is one of the most frequently c
onsidered methods for source location using sensor arrays. Among the l
ocation methods based on one-dimensional search, MUSIC has excellent p
erformance. in fact, no other one-dimensional method that may outperfo
rm MUSIC (in large samples) was known to exist Our goal here is to int
roduce such a method, called improved sequential MUSIC (IES-MUSIC), wh
ich is shown to be strictly more accurate than MUSIC (in large samples
). First, a class of sequential MUSIC estimates is introduced, which d
epend on a scalar-valued user parameter. MUSIC is shown to be a specia
l case of estimate in that class, corresponding to a value of uro for
the user parameter. Next, the optimal user parameter value, which mini
mizes the asymptotic variance of the estimation errors, is derived IES
-MUSIC is the method based on that optimal choice of the user paramete
r. Simulation results which Lend support to the theoretical findings a
re included.