remind them of correl/convol thm's; the correlation of the two in one domain becomes the product of the two in the other domain. thus A corr B in time domain gives FT(A) times FT(B) in other domain. this also leads to the identity FT (ACF (V)) = FT(V) * comp conj FT(V) = pwr spectrum illus with the horn data. gv acf_acrplot.ps talking points: 1. top panel is power spectrum: red and black similar, but not quite identical. 2. panel 2 shows that for the spike, a bif diff--P(f) from ACF goes negatave! 3. P(f) comes from FT(ACF). ACF is in panel 3, 4. it looks like mainly zeros for tau > 2 microsec, but those small numbers repeat thoussands of times and have most of the info! 4. To have a reasonably high ACF, you need small tau. in particular, the first zero is at about 0.3 microsec. The bandwidth is, say, 2 MHz or 4 MHz. (1/B) = (1/4MHz) = 0.215 microsc. FOR THE INTERFEROMETER, B=30 MHz, so first zero is at 3.3e-8 sec or 33 nsec. during this time, length traveled by light is ct = 3e8 * 3.3e-8 = 10 meters, about half our baseline. at west horizon, east telescope is delayed from west by B and at east, west from east by B, a total change of 2B, or about 35 meters, equiv to a change in time delay of (35/3e8) = 1e-7 sec, 3 times longer than tau at first zero. now, explore monochromatic waves. gv xv_vs_fx_1.ps the two P(f)s (black/red) differ. but if time interval wene to infty, there would be no leakate power and they would be identical. all is not lost for ACF method; do Hanning smoothing. makes the two estimates almost identical (green). gv xv_vs_fx_2.ps shows what we convolvew with in freq space to get rid of sinx/x. makes sense from an intuuitive stanedpoint. now use corrl thm: a convol in freq space is a multiplication in time of the two FTs in time space. The FT of the hanning spoke is 0.5 + 0.5 cos() (middle panel). this is called 'tapereing'. you trae rresolution for sidelobes. application to telescopes...to reduce spatial sidelobes. Remark on the use of 'correlators' in radio spectroscopy.