Page 240 - Electrical Engineering Dictionary

(3) automatic adjustment of the multipro- where c(k) =[i k ,j k ], such that the distance
cessing program at run time that reflects the
K
actual number of CPUs available presently. . X 2
d(A, B) == (a i(k) − b j(k) )
For instance, a DO loop with 100 itera-
k=1
tions is automatically scheduled as 2 blocks
with 50 iterations on a two-processor sys- is minimum. The optimization must take
tem, as 10 blocks with 10 iterations on a place under the following conditions:
ten-processor system, and as one block on a 1. monotonic condition
single-processor machine. This enables one i(k) ≥ i(k − 1) and j(k) ≥ j(k − 1)
to run multiprocessor programs on single- 2. boundary conditions
processor computers. i(1) = k(1) = 1
i(K) = M
dynamic simulation See direct dynamics. j(K) = N
3. non-skip condition i(k)−i(k−1) ≤ 1 and
j(k) − j(k − 1) ≤ 1
dynamic stability a measure of a power 4. efficiency condition
system to return to a pre-disturbance steady- |i(k) − j(k)| <Q
state condition following a disturbance. The solution of this problem can be ob-
tained by Belmann’s dynamic programming.
dynamic system See static system. The algorithm that produces the optimal tem-
plate alignment is referred to as dynamical
dynamic time division multiple access time warping (DTW).
(D-TDMA) time division multiple access
scheme in which the channels are assigned dynamic time warping (DTW) a recog-
dynamically. See also time division multiple nition technique based on nonlinear time
access. alignment of unknown utterances with ref-
erence templates.
dynamic time warping in problems of
dynamical linear nonstationary
temporal pattern recognition, each exemplar
continuous-time finite-dimensional sys-
can be regarded as a sequence of vectors.
tem a system described by the linear
The process of pattern matching requires to
ordinary differential state-equation
carry out an optimal alignment of the vectors
composing the sequences so as to minimize 0
x (t) = A(t)x(t) + B(t)u(t)
a proper distance. For example, in automatic
speech recognition, the problem of isolated
and the linear algebraic output equation
word recognition requires producing an op-
timal alignment between the incoming word y(t) = C(t)x(t) + D(t)u(t)
to be classified and a reference template. Let
where
n
A = [a 1 ,..., a M ] x(t) ∈ R
B = [b 1 ,..., b N ]
is the state vector,
p
be two sequences of vectors (a i , b i ∈ R ) m
u(t) ∈ R
that must be aligned optimally. Formally, de-
termining the optimal alignment consists of
is the input vector,
finding a warping function
y(t) ∈ R q
C = c 1 ,...,c K

c
2000 by CRC Press LLC