Binary variational (biv) CSI coding
In some example embodiments, there may be provided a method that includes receiving, by a machine learning encoder as part of a training phase, channel state information as data samples; generating, by the machine learning encoder, a latent variable comprising a log likelihood ratio value representation for the channel state information, wherein the latent variable provides a lower dimension binary representation when compared to the received channel state information to enable compression of the received channel state information; generating, by the binary sampler, a binary coding value representation of the latent variable, wherein the binary coding value converts the latent variable to a binary form; and generating, by the machine learning decoder, a reconstructed channel state information, wherein the generating is based in part on the binary coding value representation of the latent variable generated by the binary sampler. Related systems, methods, and articles of manufacture are also disclosed.
1 . A method for predicting using a machine learning based channel state information coding system, comprising:
providing, to a machine learning encoder comprised at a user equipment, channel state information to generate a latent variable representation of the channel state information comprising a log-likelihood ratio vector of log-likelihood ratio values;
performing binary sampling, by a binary sampler comprised at the user equipment, to generate a random binary vector representing the channel state information from the log-likelihood ratio vector wherein the binary sampling converts the log-likelihood ratio vector into the random binary vector based on a sign function that uses a log-likelihood ratio value and a random variable as input; and
sending, by the user equipment, a measurement report to a base station, wherein the measurement report includes the random binary vector representing the channel state information to enable the base station to decode the random binary vector, using a machine learning decoder, in order to reconstruct the channel state information.
2 . The method of claim 1 , wherein the machine learning decoder of the base station is comprised in the machine learning based channel state information coding system.
3 . The method of claim 1 , wherein the machine learning encoder and the machine learning decoder are trained using an objective function, wherein the objective function includes a variational probability for the machine learning encoder, a posteriori probability for the machine learning decoder, and a priori probability of the random binary vector.
4 . The method of claim 3 , wherein the objective function comprises a sum of the posteriori probability for the machine learning decoder and the priori probability of the random binary vector less the variational probability for the machine learning encoder.
5 . The method of claim 3 , wherein the machine learning encoder and the machine learning decoder are trained using backpropagation and the objective function to determine one or more weights of the machine learning encoder and the machine learning decoder, and wherein the objective function uses at least one input comprising the random binary vector.
6 . The method of claim 3 , wherein the machine learning encoder and the machine learning decoder are comprised in a variational auto encoder trained with the objective function.
7 . The method of claim 3 , wherein the objective function may be represented by the following equation:
L
˜
θ
,
ϕ
(
x
)
=
log
p
θ
(
x
❘
"\[LeftBracketingBar]"
z
)
+
log
p
θ
(
z
)
-
log
q
ϕ
(
z
❘
"\[LeftBracketingBar]"
x
)
,
wherein ϕ indicates the machine learning encoder, θ indicates the machine learning decoder, {tilde over (L)} θ, ϕ (x) represents the objective function, p θ (x|z) is the posteriori probability for the machine learning decoder, q ϕ (z|x) is the variational probability for the machine learning encoder, and p θ (z) is the priori probability of the random binary vector.
8 . The method of claim 1 , wherein the machine learning encoder is comprised in an artificial intelligence engine at the user equipment.
9 . The method of claim 1 , wherein the random variable comprises a Gaussian random variable.
10 . The method of claim 1 , wherein the log-likelihood ratio values output by the machine learning encoder represent a Bernoulli distribution sampled by the binary sampler.
11 . An apparatus comprising user equipment, the apparatus comprising:
at least one processor; and
at least one memory including computer program code, the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus to at least:
provide, to a machine learning encoder comprised at the user equipment, channel state information to generate a latent variable representation of the channel state information comprising a log-likelihood ratio vector of log-likelihood ratio values;
perform binary sampling, by a binary sampler comprised at the user equipment, to generate a random binary vector representing the channel state information from the log-likelihood ratio vector, wherein the binary sampler converts the log-likelihood ratio vector into the random binary vector based on a sign function that uses a log-likelihood ratio value and a random variable as input; and
send a measurement report to a base station, wherein the measurement report includes the random binary vector representing the channel state information to enable the base station to decode the random binary vector, using a machine learning decoder, in order to reconstruct the channel state information.
12 . The apparatus of claim 11 , wherein the machine learning encoder and the machine learning decoder are trained using an objective function, wherein the objective function includes a variational probability for the machine learning encoder, a posteriori probability for the machine learning decoder, and a priori probability of the random binary vector.
13 . The apparatus of claim 12 , wherein the objective function comprises a sum of the posteriori probability for the machine learning decoder and the priori probability of the random binary vector less the variational probability for the machine learning encoder.
14 . The apparatus of claim 12 , wherein the machine learning encoder and the machine learning decoder are trained using backpropagation and the objective function to determine one or more weights of the machine learning encoder and the machine learning decoder, and wherein the objective function uses at least one input comprising the random binary vector.
15 . The apparatus of claim 12 , wherein the machine learning encoder and the machine learning decoder are comprised in a variational auto encoder trained with the objective function.
16 . The apparatus of claim 12 , wherein the objective function may be represented by the following equation:
L
˜
θ
,
ϕ
(
x
)
=
log
p
θ
(
x
❘
"\[LeftBracketingBar]"
z
)
+
log
p
θ
(
z
)
-
log
q
ϕ
(
z
❘
"\[LeftBracketingBar]"
x
)
,
wherein ϕ indicates the machine learning encoder, θ indicates the machine learning decoder, {tilde over (L)} θ, ϕ (x) represents the objective function, p θ (x|z) is the posteriori probability for the machine learning decoder, q ϕ (z|x) is the variational probability for the machine learning encoder, and p θ (z) is the priori probability of the random binary vector.
17 . The apparatus of claim 11 , wherein the machine learning encoder is comprised in an artificial intelligence engine at the user equipment.
18 . The apparatus of claim 11 , wherein the random variable comprises a Gaussian random variable.
19 . The apparatus of claim 11 , wherein the log-likelihood ratio values output by the machine learning encoder represent a Bernoulli distribution sampled by the binary sampler.