Using example x99, with 2 parabolic tracks, a varying number of missed
target events, a varying number of false detections (Poisson distribution),
a penalty for unused measurements, plus apriori probabilities in the
cost function (P(missed target)=0.2, P(false detection)=Poisson with rate 2),
w(K) is the total cost of the best paths for K=1..4 (using Kn=0.01, S2a=2, L=4):

  w =

    162.81  135.23  161.13  214.35

The minimum cost occurs for K=2, which agrees with the actual input
data used for the simulation (2 tracks).

Examining the best paths for K=4:

[combos(:,path_index(:,1))', n1', n2', n_false_detect']

  	1  2  3  4	  1  1  2
  	1  2  4  7	  1  1  4
  	1  2  7  7	  1  1  1
  	1  2  7  7	  1  1  0
  	1  2  7  7	  1  1  1
  	1  7  7  7	  1  0  1
  	1  2  7  7	  1  1  1
  	1  7  7  7	  1  0  1
  	1  2  7  7	  1  1  2
  	1  2  4  7	  1  1  4
  	1  2  7  7	  1  1  1
  	1  2  7  7	  1  1  2
  	1  7  7  7	  0  1  1
  	1  4  7  7	  1  0  3
  	1  7  7  7	  1  0  2
  	1  7  7  7	  1  1  2
  	1  7  7  7	  1  1  1
  	1  7  7  7	  1  1  1
  	1  7  7  7	  1  1  2
  	1  4  7  7	  1  1  2
  	1  7  7  7	  1  0  1

The last column of the above table shows the number of false detections
generated per unit time.  The n1 and n2 columns show the number of
measurements used for the true tracks.  These were generated randomly,
with P(missed target)=0.2, and in this case one sample for target #1 and
five samples for target #2 were generated as missed detections.

The first four columns show the measurement indices used in the best set of
four tracks as determined by the K-path L-best Viterbi algorithm.  In these
columns, the number 7 represents a missed target event detection.  Numbers
1 to 6 represent selections of the measurements, with 1 and 2 corresponding
to real target measurements (except for rows where n1 or n2 is zero) and 3
to 6 corresponding to false detections.  Row 1 is for time=1, row 2 for
time=2, etc.

The first track (first column) used measurement 1 each time.  In the row
where n1=0, measurement 1 was actually a value from the second generated
track, but the two parabolic tracks have similar values in the middle of
the simulation where they are tangent, so this is reasonable.

The second track (second column) correctly used measurement 2, except
when n2=0 where it correctly used the missed target event, for the first
half of the simulation.  Then, after the two sequential missed target events,
it looses the track and continues to choose missed target events and one
false detection (measurement 4 in next to last row).  The two sequential
missed target events for track #2 occur just where the parabolic track
starts to accelerate upwards.

The third and fourth tracks (third and fourth column) consist mostly of
missed target events, a clear indication that they are false and the
number of true tracks is 2.