Method and apparatus for machine-learning-based resource allocation in wireless communication system
View Patent ↗The present disclosure relates to a machine learning-based resource allocation method and device in a wireless communication system. A method of performing resource allocation in a wireless communication system according to an embodiment of the present disclosure may include acquiring DNN output information through a deep neural network (DNN) based on input H corresponding to wireless channel-related information; acquiring resource allocation matrix X having a form of a permutation matrix through an approximation operation to a doubly stochastic matrix based on input A corresponding to the DNN output information; and allocating M resource elements belonging to a set of first type resources to N resource elements belonging to a set of second type resources based on the resource allocation matrix.
1 . A method of performing resource allocation in a wireless communication system, the method comprising:
acquiring DNN output matrix A through a deep neural network (DNN) based on input matrix H;
directly acquiring resource allocation matrix X having a form of a permutation matrix through an approximation operation to a doubly stochastic matrix with applying the DNN output matrix A as an input; and
allocating M resource elements belonging to a set of first type resources to N resource elements belonging to the set of second type resources based on the resource allocation matrix X,
wherein an input vector h ii of the input matrix H corresponds to a radio wave environment or a connection state between an i-th first type resource of the first type resources and a i-th second type resource of the second resources,
wherein the DNN output matrix A corresponds to N×N-sized matrix reshaped from a DNN output vector a,
wherein, for M=N,
∑
i
=
1
N
x
ij
=
1
,
∀
j
,
and
∑
j
=
1
N
x
ij
=
1
,
∀
i
,
where x jj (i∈M, i∈N) corresponds to a binary optimization variable or a discrete allocation variable having a value of 0 or 1, and
wherein, for M>N or N>M.
∑
i
=
1
N
x
ij
=
1
,
∀
j
,
and
∑
j
=
1
M
x
ij
≤
1
,
∀
i
.
2 . The method according to claim 1 , wherein:
the approximation operation has a limited number of repetitions.
3 . The method according to claim 2 , wherein:
the approximation operation is performed consecutively K times and the consecutively performed operation is repeated L/K times.
4 . The method according to claim 3 , wherein:
the approximation operation includes applying a constant representing an approximation degree to the DNN output information.
5 . The method according to claim 4 , wherein:
the approximation operation is S(τA),
τ is the constant representing the approximation degree.
6 . The method according to claim 5 , wherein:
the approximation operation is applied as an activation function of a last layer of a neural network deriving the resource allocation matrix X.
7 . The method according to claim 6 , wherein:
a process of acquiring the resource allocation matrix X is performed with an optimization process of other element of the wireless communication system.
8 . The method according to claim 5 , wherein for M=N:
S
(
A
)
=
arg
max
D
∈
𝒟
n
tr
(
A
T
D
)
,
D n corresponds to the set of n×n-sized doubly stochastic matrixes,
tr( ) refers to a diagonal sum,
S( ) includes repetition of row-wise normalization and column-wise normalization,
D is the doubly stochastic matrix,
d ij , elements of D, satisfies
∑
i
=
i
n
d
i
j
=
1
,
∑
j
=
1
n
d
ij
=
1
,
0
≤
d
i
j
≤
1.
9 . The method according to claim 8 , wherein:
S( ) is defined by
S
0
(
A
)
=
△
exp
(
A
)
,
S
m
(
A
)
=
C
(
R
(
S
m
-
1
(
A
)
)
)
,
R( ) corresponds to the row-wise normalization,
C( ) corresponds to the column-wise normalization,
they are defined by
R
(
A
)
=
△
{
a
~
ij
,
∀
i
,
j
}
,
a
~
ij
=
a
ij
∑
k
=
1
n
a
ik
,
C
(
A
)
=
△
{
a
^
ij
,
∀
i
,
j
}
,
a
^
ij
=
a
ij
∑
k
=
1
n
a
kj
.
10 . The method according to claim 5 , wherein for M>N or N>M:
a matrix excluding N−M or M−N columns of the resource allocation matrix X is defined as a modified resource allocation matrix,
the resource allocation step is performed based on the modified resource allocation matrix.
11 . A device performing resource allocation in a wireless communication system, the device comprising:
a memory; and
a processor,
wherein the processor is configured to:
acquire DNN output matrix A through a deep neural network (DNN) based on input matrix H;
directly acquire resource allocation matrix X having a form of a permutation matrix through an approximation operation to a doubly stochastic matrix with applying the DNN output matrix A as an input; and
allocate M resource elements belonging to a set of first type resources to N resource elements belonging to the set of second type resources based on the resource allocation matrix X,
wherein an input vector h ii of the input matrix H corresponds to a radio wave environment or a connection state between an i-th first type resource of the first type resources and a i-th second type resource of the second type resources,
wherein the DNN or put matrix A corresponds to a N×N-sized matrix reshaped from a DNN output vector a,
wherein, for M=N,
∑
i
=
1
N
x
ij
=
1
,
∀
j
,
and
∑
i
=
1
N
x
ij
=
1
,
∀
i
,
where x jj (i∈M, j∈N) correspond to a binary optimization variable or a discrete allocation variable having a value of 0 or 1, and
wherein, for N<M or M<N,
∑
i
=
1
N
x
ij
=
1
,
∀
j
,
and
∑
j
=
1
M
x
ij
≤
1
,
∀
i
.