Analysis of two streamer mini-sparker data
Step 1
 lsd Solstice2014_L43.sgy | head
      SHOT   TR       RP    TR ID  RANGE DELAY NSAMPS    SI   YR DAY HR MIN SEC
      5672     0       0     0  1      0     0  10000   100 2014 133 11   3  21
      5672     1       0     0  1      0     0  10000   100 2014 133 11   3  21
         2     0       0     0  1      0     0  10000   100 2014 133 11   3  21
         2     1       0     0  1      0     0  10000   100 2014 133 11   3  21
         3     0       0     0  1      0     0  10000   100 2014 133 11   5  39
         3     1       0     0  1      0     0  10000   100 2014 133 11   5  39
         4     0       0     0  1      0     0  10000   100 2014 133 11   5  40
         4     1       0     0  1      0     0  10000   100 2014 133 11   5  40
         5     0       0     0  1      0     0  10000   100 2014 133 11   5  41

Trace 0 is not honored by most systems.  Sioseis also assumes shot numbers are
monotonically increasing when the SEG Rev 1 "fixed length" flag is set.
REWRITE the file using diskin renum 1 retrac 1

Step 2
   Look at some traces from shots 101-120
Plot some trace 1s.
Plot of raw shots 101-150, trace 1
Plot of filtered  shots 101-150, trace 1
Plot of raw shots 101-150, trace 2
Plot of filtered shots 101-150, trace 2
Plot of filtered shots 101-120, traces 1 & 2
traces 1 and 2 are reversed polarity from each other.
Plot of weighted filtered shots 101-120, traces 1 & 2
notice that trace 2 seems to be time shifted
Plot of filtered shots 101-120, traces 1 & 2, trace 1 reversed polarity, trace 2 shifted 1 sample
traces 1 and two have different frequency - trace 2 have much more low frequency
Plot of filtered (120x500) shots 101-120, traces 1 & 2, trace 1 reversed polarity, trace 2 shifted 1 sample
notice that traces 1 & 2 have different amplitude
Step 2
   Look at some traces from shots 4101-4120
Plot of weighted (tsp 1 -1.3) shifted (tsp 2 -.0001) filtered (120x500) shots 4101-4120, traces 1 & 2 
Trace 2 is not shifted enough.  Trial and error shows .0006s shift might be best.
Plot of weighted (tsp 1 -1.3) shifted (tsp 2 -.0006) filtered (120x500) shots 4101-4120, traces 1 & 2 
notice that the traces 1&2 line up pretty well and are more similar than before.
Theory on recording problems
1)  Streamer 1 (trace 1) is connected incorrectly which resulted in:
    reverse polarity
    lower amplitude
    occasional spikes (that get anti-alias filtered)
    more high frequency noise
2)  Streamer 1 is towed ~1 meter deeper than streamer 2 which results in streamer 1
    returns (in two-way-traveltine) coming in .0006s before streamer 2. 
    (1m / 1500m/s / 2 = .000666666s)
3)  There are occasional "misfires" such as record 4103.
Compare stack vs trace 1 vs trace 2
Plot 1 of stack of trace 1 and trace 2 with a two trace record mix.
    procs diskin weight shift stack mix filter mute t2d plot end
    weight twp 1 -1.3 shift tsp 2 -.0006 filter pass 120 500 
    mute ttp 1 .2 mix type 3 weight 1 1 t2d osi .1 sdepth 0 edepth 500 vtp 1500 0 
Plot 2 of Streamer 1 with 2 trace record mix.
    procs diskin weight mix filter mute t2d plot end
    weight twp 1 -1.3 filter pass 120 500 mute tttp 1 .2 mix type 3 weight 1 1 
    t2d osi .1 sdepth 0 edepth 500 vtp 1500 0
Plot 3 of Streamer 2 with 2 trace record mix.
    procs diskin weight mix filter mute t2d plot end
    filter pass 120 500 mute tttp 1 .2 mix type 3 weight 1 1 
    t2d osi .1 sdepth 0 edepth 500 vtp 1500 0
Conclusions so far
    Use trace/streamer 2 and ignore streamer 1.  While the corrected streamer 1 
stacked with shifted streamer 2 has slightly better SNR (signal-to-noise ratio),
that may not hold on other seismic lines or when one of the streamers changes depth.
The connector problem may disappear when/if the streamer is deployed again.  
Likewise, the streamer depths may change when redeployed.
Other possible improvements (to be investigated)
1)  Find a water-bottom picking method so the process swell can be used (see day 133 1215z)
2)  Find an automatic bad shot picker - e.g. shot 4103.
3)  Find an automatic "spike" detector and kill the shot - e.g. shot 4113 trace 1
More tries at improvement
1)  Bad shot/trace killer.
    The plot from step 2 shows shot 4103 is likely a "misfire" and doesn't contain
    much useful seismic information.  Process tredit was used to kill these bad traces.
    The sioseis screen output shows what traces were killed and the parameters
    used, with explanations of the tredit parameter choice.  The (plot) doesn't
    show much improvement from plot 2 above.
2)  Englarge this plot to examine the water bottom and the
    effects of the sea swell.  Pick the water bottom for process swell and hanging mutes.  
    Also, enlarged plot the stacked mixed 
    shots 102-120 and 4102-4120 for reference.
    a) Process wbt, solrat/ses/sel doesn't work well because the shot to shot source variation
       is too great.  Not only is the shape of the waveform different, but the amplitudes
       and energy are quite different.  See the picks.
    b) Using process prout to examine the amplitudes (stdout)
       and wbt, thres .2, doesn't work well either.
    c) Using process wbt with  peak pos sepp .25 .3 (stdout) shows that the peak 
       varies a bit because of the shot waveform differences.  Parameter track .002 should help.
       That doesn't help.
    d) Use the stack without the record mix.  
       wbt  pass 250 500 track .002 peak pos sepp .25 .35 end  doesn't work in the section of
          shots 1600 - 2500 because of the very strong reflector at .32s
    e) Re-examining the amplitudes (stdout) along with the plot of 102-120,
       wbt thres .1 track .002 detected the beginning of the waveform (a negative number),
       but lost track at the diffraction around shot 1350.  track .003 is better.
    f) Examining the enlarged plots, the period of the swell is 12 shots.  Remembering
       that the stacked mixed data is compressed by a factor of 2, so set process swell
       parameter n to 6.  Check the wbt swell by redoing the "enlarged plot".
       Looks good.  Compare the stack plot with tredit and swell with Plot 1 above.
       Script was used to screate the stack with swell.
       Enlarge the plots and flip between the two browser tabs to see the improvement!
       Doing wbt swell before the two trace record mix would be better.