The null spectrum defined in multi image subspace algorithm (MISA) for sour
ce localization in shallow water possesses an interesting property, that is
, when it is averaged over the signal bandwidth, its mean is found to clust
er between 0.8 and 0.9, except at the actual source position where it is cl
ose to zero. Thus, the frequency-averaged null spectrum or its inverse is f
ound to be an effective quantity for source localization. The signal peak t
o maximum side lobe ratio is found to increase as a square root of the numb
er of frequency bins over which it is averaged. An approximate theoretical
analysis has been carried out. The theoretical results broadly support the
outcome of the numerical experiments. Further, we have shown that for broad
band averaging to succeed the segment length must be much greater than the
largest delay between the direct path and the furthest significant multipat
h. (C) 1999 Elsevier Science B.V. All rights reserved.