Information Technology Reference
In-Depth Information
the third part of one packet. Every packet received can recompute the hash value of
the root using the third part, so the different group with fewer ones will be discarded.
3.2
The Estimate Model Using the Markov Chain
We apply the Markov chain to estimate the incoming packet error rate based on the
previous ones. As we all know, the Markov chain describes a kind of state sequence
that every state in the sequence depends on the previous finite ones according to the
state transition matrix constituted by the transition probabilities. So what we want to
get is the transition matrix of the data packet error rate. At the same time, we adjust
the time delay based on the feedback values.
The Adjustment of Block Size Parameter
We can know from the IDA that the more packets included in one block, the smaller
overhead each packet will take. On the other hand, the time delay will increase with n .
So the balance between these factors is necessary.
Here we define a threshold value of the time delay as t , as well as a threshold limit
time t 0 . Only if the average time delay in one period exceeds the threshold value t , the
size n will be decreased. And if the time delay keeps under the threshold value for t 0
time, then we will increase the size n .
Then we also need to consider the overhead with the value n . It should be con-
trolled so that the overhead per packet would not be too big. In our experiments, the n
should be no more than 128 considering the limit of the experiment environment.
The Estimate of Packet Error Rate
First of all, we define k states which present the default values of packet error rate as
O 1 , O 2 ,…, O k (0 ≤ O i < 1 and 1 ≤ i k ). And the k values are set as the coordinates of
the transition matrix, in which the a ij (1 ≤ i , j k ) presents the probability count of
transition between the rates as (1). For example, a 2 k means the situation that the pre-
vious packet error rate is O 2 and the next one is O k appears a 2 k times in this period.
O
O
O
aa
,
,...,
a
O
11
12
1
k
aa
,
,...,
a
O
(1)
21
22
2
k
aa
,
,...,
a
O
k
1
k
2
kk
What should be noticed is that k must be an integer in the range of (0, n ]. The bigger
the k is, the more accurate the matrix is but the estimate may be inaccurate with too
many a ij = 0 in the matrix. The smaller the k is, the more accurate the estimate is.
However, the estimate may be limit within several values. So the k is better fixed in
the range of [8, 32] to get a good result in the following experiments. And the values
Search WWH ::




Custom Search