Team # 32738
Abstract
For problem 1, in order to simulate the communication behavior of the distributed
wireless broadcast networks, the basic input parameters of the model related to the
environment are first determined. According to whether the communication node is in
a low-speed continuous motion state, we established Static Distributed Network (SDN)
model and a Mobile Distributed Network (MDN) model through MATLAB, and
established a calculation algorithm for derived input parameters. We tested the models.
Secondly, we constructed as many description parameters as possible based on
model output parameters. We discussed their practical significance. We set an
appropriate range of the input parameters and generated 5000 samples by the MDN
model. We use SPSS to perform Factor Analysis basing on these samples. After
parameter testing, estimating the factor load matrix based on the Principal Component
Analysis (PCA) method, and rotating factor based on the maximum variance method,
we obtained four different levels of evaluation indicators describing the overall
communication efficiency of the network, which are "number of conflicts or proportion
of conflicts", "average frequency of conflicts or their range of change", "is or not there
are small amount of errors in a high-quality network", and "signal transmission failure
due to distance". We also give the formula for calculating these factor score and
specified their meanings and which situation they should be considered detaily.
For problem 2, in order to reveal the optimal retransmission time in essence, we
observed the probability density function of the time of signal conflicts over time.We
selected suitable data, and smoothed them by using 12-degree Polynomial Curve
Fitting in order to obtain the conflict time in samples. By using it, we get the probability
density’s local minimum points. We combined them with input parameters, and
recorded 13762 samples in total.
Then,we constructed a Radial Basis Function Neural Network (RBFNN) model
based on Levenberg-Marquardt filtering using MATLAB. We summarized the reasons
for the low R-value. We one-sided samples, and construct a RBF model again. Then we
get a better result. We analyzed and tested our model. And we output the result of
RBFNN in a MATLAB function, in order to facilitate real-time calculations.
Additionly, we adjusted some basic input parameters, and analysed the stability of
the MDN model.
Key Words: MATLAB; factor analysis; polynomial curve fitting; RBFNN
