We propose a benchmark to test the the validity of different
implementations of the LB method for wave simulation. All the
possible parameters are fixed to a specific value and the
resulting image may be compared. The parameters are explained
elsewhere.
The layout file in (here irrelevant) natural unit. It
contains 3 buildings representing the 3 letters T L M.
The parameters may be fixed as following:
T = 8 (period of the source)
A = 0.1529 (amplitude of the source)
N = 256 (dimension of the array Nx and Ny)
L = 24 (propagation last for 24*8*2=384 iterations. Intergration is
performed during the last two period, namely for i=)
[209,209] (coordinates of the source in natural unit.
On the array the source appears at position [128,128])
0.9 (scale: one pixel represents 0.9 natural unit)
15.0 (angle of mapping between natural and array coordinates)
-0.75 (the relexion coefficient applied on the wall. Here, the
wall do not transmit flux inside the buildings.)
C0=0, C1=0.436, C2=0.8 (the adsorbing coefficients on the boundaries.
Here the boundaries are included in the original array)
The resulting image (see below) is a color PGM (raw) file using a
pseudo color map with 256 entries (How to get the RGB values for an
entry between 0 and 256 is explained here). The
colors represent the pathloss: 20*LOG10(amplitude) between the two
following extrema: -100 dB (deep blue) and -1 dB (red).
On our Sparc 10, this calculation takes about 45 seconds to complete. the
memory used is 4 Mbytes. On a IBM RS 6000 we need about 20 seconds.
