Electronic device and operation method thereof
Provided are an electronic device capable of learning a log likelihood ratio (LLR) scaling factor distribution, and an operation method of the electronic device. The operation method of the electronic device includes receiving training environment information, obtaining a measurement log likelihood ratio (LLR) distribution, based on the training environment information, obtaining an inter-distribution divergence value, based on a reference LLR distribution and the measurement LLR distribution, and obtaining an LLR scaling factor distribution by converting the inter-distribution divergence value into a probability value.
1 . An operation method of an electronic device, the operation method comprising:
receiving training environment information;
obtaining a measurement log likelihood ratio (LLR) distribution, based on the training environment information;
obtaining an inter-distribution divergence value, based on a reference LLR distribution and the measurement LLR distribution;
obtaining an LLR scaling factor distribution by converting the inter-distribution divergence value into a probability value;
receiving communication environment information;
selecting a first LLR scaling factor corresponding to the communication environment information;
obtaining an output LLR distribution, based on input data and the first LLR scaling factor; and
obtaining decoding data by decoding the output LLR distribution.
2 . The operation method of claim 1 , wherein the selecting comprises selecting the first LLR scaling factor corresponding to a largest probability value among probability values included in the LLR scaling factor distribution.
3 . The operation method of claim 1 , wherein the inter-distribution divergence value, f, is calculated according to equation below
f
(
k
❘
"\[LeftBracketingBar]"
TEI
,
y
j
)
=
{
P
y
j
,
TEI
(
k
)
ln
P
y
jj
,
TEI
(
k
)
P
(
Δ
k
,
Q
)
,
if
P
y
j
,
TEI
(
k
)
>
P
(
Δ
k
,
Q
)
,
k
=
1
or
2
n
+
1
P
(
Δ
k
,
Q
)
ln
P
(
Δ
k
,
Q
)
P
y
j
,
TEI
(
k
)
,
otherwise
,
wherein P y j, TEI (k) is a probability distribution of the measurement LLR distribution, TEI indicates training environment information, y j indicates a measurement LLR value for a j-th training LLR scaling factor being selected, and
wherein Δ k indicates a number of k-th quantized LLR values, and Q indicates a reference LLR distribution.
4 . The operation method of claim 1 , wherein the probability value is calculated according to equation below
P
(
y
j
,
TEI
)
=
exp
(
E
(
y
j
,
TEI
)
)
∑
j
=
1
c
exp
(
E
(
y
j
,
TEI
)
)
,
wherein E(y j , TEL) is an estimator function, TEI indicates training environment information, y j indicates a measurement LLR value for a j-th training LLR scaling factor being selected, and c indicates a number of configurable training LLR scaling factors.
5 . The operation method of claim 1 , wherein the training environment information comprises at least one of a Signal to Interference plus Noise Ratio (SINR), a modulation order (MO), new transmission or non-transmission, and a number of layers.
6 . The operation method of claim 1 , wherein a probability distribution of the reference LLR distribution is calculated according to equation below, for which Δ k indicates a number of k-th quantized LLR values, ={Δ 1 , Δ 2 , . . . , Δ 2n+1 } indicates a distribution of quantized LLR values quantized to 2n+1 quantized LLR values, Q indicates a reference LLR distribution, and τ k indicates 2 −k :
P
(
Δ
k
,
Q
)
=
{
∑
k
∈
ℒ
Δ
k
-
1
2
∑
k
∈
ℒ
Δ
k
if
k
=
1
or
2
n
+
1
,
τ
k
-
1
P
(
Δ
1
,
Q
)
if
2
≤
k
<
n
+
1
,
τ
2
n
+
1
-
k
P
(
Δ
1
,
Q
)
otherwise
.
7 . The operation method of claim 1 , wherein a probability distribution of the measurement LLR distribution is calculated according to equation below
P
y
j
,
TEI
(
k
)
=
LLR
k
,
j
LLR
total
,
j
,
1
≤
k
≤
2
n
+
1
,
wherein TEI indicates training environment information, y j indicates a measurement LLR value for a j-th training LLR scaling factor being selected, LLR k,j indicates the number of k-th quantized LLR values, and LLR total,j indicates the total number of first through (2n+1) th quantized LLR values.
8 . An electronic device comprising:
a log likelihood ratio (LLR) scaling factor selector configured to receive communication environment information and select a first LLR scaling factor corresponding to the communication environment information;
a symbol detector configured to obtain an output LLR distribution, based on input data and the first LLR scaling factor,
wherein the LLR scaling factor selector receives training environment information, obtains a measurement LLR distribution based on the training environment information, obtains an inter-distribution divergence value based on a reference LLR distribution and the measurement LLR distribution, and converts the inter-distribution divergence value into a probability value to obtain an LLR scaling factor distribution; and
a decoder configured to obtain decoding data by decoding the output LLR distribution.
9 . The electronic device of claim 8 , wherein the LLR scaling factor selector selects the first LLR scaling factor corresponding to a largest probability value among probability values included in the LLR scaling factor distribution.
10 . The electronic device of claim 8 , wherein the inter-distribution divergence value, f, is calculated according to equation below
f
(
k
❘
"\[LeftBracketingBar]"
TEI
,
y
j
)
=
{
P
y
j
,
TEI
(
k
)
ln
P
y
jj
,
TEI
(
k
)
P
(
Δ
k
,
Q
)
,
if
P
y
j
,
TEI
(
k
)
>
P
(
Δ
k
,
Q
)
,
k
=
1
or
2
n
+
1
P
(
Δ
k
,
Q
)
ln
P
(
Δ
k
,
Q
)
P
y
j
,
TEI
(
k
)
,
otherwise
,
wherein P y j, TEI (k) is a probability distribution of the measurement LLR distribution, TEI indicates training environment information, y j indicates a measurement LLR value for a j-th training LLR scaling factor being selected, and
wherein Δ k indicates a number of k-th quantized LLR values, and Q indicates a reference LLR distribution.
11 . The electronic device of claim 8 , wherein the probability value is calculated according to equation below
P
(
y
j
,
TEI
)
=
exp
(
E
(
y
j
,
TEI
)
)
∑
j
=
1
c
exp
(
E
(
y
j
,
TEI
)
)
,
wherein E(y j , TEI) is an estimator function, TEI indicates training environment information, y j indicates a measurement LLR value for a j-th training LLR scaling factor being selected, and c indicates a number of configurable training LLR scaling factors.
12 . The electronic device of claim 8 , wherein the training environment information comprises at least one of a Signal to Interference plus Noise Ratio (SINR), a modulation order (MO), new transmission or non-transmission, and a number of layers.
13 . The electronic device of claim 8 , wherein a probability distribution of the reference LLR distribution is calculated according to equation below, for which Δ k indicates a number of k-th quantized LLR values, ={Δ 1 , Δ 2 , . . . , Δ 2n+1 } indicates a distribution of quantized LLR values quantized to 2n+1 quantized LLR values, Q indicates a reference LLR distribution, and τ k indicates 2 −k :
P
(
Δ
k
,
Q
)
=
{
∑
k
∈
ℒ
Δ
k
-
1
2
∑
k
∈
ℒ
Δ
k
if
k
=
1
or
2
n
+
1
,
τ
k
-
1
P
(
Δ
1
,
Q
)
if
2
≤
k
<
n
+
1
,
τ
2
n
+
1
-
k
P
(
Δ
1
,
Q
)
otherwise
.
14 . The electronic device of claim 8 , wherein a probability distribution of the measurement LLR distribution is calculated according to equation below
P
y
j
,
TEI
(
k
)
=
LLR
k
,
j
LLR
total
,
j
,
1
≤
k
≤
2
n
+
1
,
wherein TEI indicates training environment information, y j indicates a measurement LLR value for a j-th training LLR scaling factor being selected, LLR k,j indicates a number of k-th quantized LLR values, and LLR total,j indicates a total number of first through (2n+1) th quantized LLR values.
15 . The electronic device of claim 8 , wherein the LLR scaling factor selector is a Bayes classifier using Gaussian Naive Bayes.
16 . A non-transitory computer readable storage medium storing instructions configured to cause a processor, when executing the instructions, to perform log likelihood ratio (LLR) scaling factor distribution learning, the LLR scaling factor distribution learning comprising:
receiving training environment information;
obtaining a measurement LLR distribution, based on the training environment information;
obtaining an inter-distribution divergence value, based on a reference LLR distribution and the measurement LLR distribution; and
obtaining an LLR scaling factor distribution by converting the inter-distribution divergence value into a probability value;
receiving communication environment information;
selecting a first LLR scaling factor corresponding to the communication environment information;
obtaining an output LLR distribution, based on input data and the first LLR scaling factor; and
obtaining decoding data by decoding the output LLR distribution.
17 . The non-transitory computer readable storage medium of claim 16 , wherein the training environment information comprises at least one of a Signal to Interference plus Noise Ratio (SINR), a modulation order (MO), new transmission or non-transmission, and a number of layers.