Changed guessed Q from 5 to 1, which gives about the right breadth.

I'm dubious about the fit quality though.  Looking at the fit.log
output after looking at the 20080919_cantA_tweak.vib data again reveals
  After 4 iterations the fit converged.
  final sum of squares of residuals : 9.6001e+08
  rel. change during last iteration : -2.46003e-07
for the first curve, and similar residuals (and exactly 4 iterations)
for the other curves.  Hmm.  I should probably explicitly show the range
of frequency space I'm fitting in.

diff --git a/calibcant/vib_analyze.py b/calibcant/vib_analyze.py
index 6fb16154b7766ad9d6c997437df656f02ad95e7e..cfa2c3e90e45663f409010cf18e1b9ef80359ac6 100755 (executable)
--- a/calibcant/vib_analyze.py
+++ b/calibcant/vib_analyze.py
@@ -157,8 +157,8 @@ def fit_psd(freq_axis, psd_data, minFreq=500, maxFreq=7000) :
     # peak height    = C / (3 x_res^4 + A^4)
     # Q = A/B
     #
-    # guess Q = 5, and adjust from there
-    Q_guess = 5
+    # guess Q = 1, and adjust from there
+    Q_guess = 1
     # so x_res^2 = B^2 Q^2 - B^2/2 = (Q^2-1/2)B^2 
     #    B = x_res / sqrt(Q^2-1/2)
     if config.TEXT_VERBOSE :