Apparatus and method for controlling non-linear effect of power amplifier
Embodiments herein disclose a method for controlling a non-linear effect of a power amplifier by an apparatus. The method includes acquiring an input data of the power amplifier of the apparatus and an output data of the power amplifier. Further, the method includes determining an inverse function using a neural network. The inverse function maps normalized output data of the PA to the input data of the PA, where the neural network comprises at least one sub-network for at least one memory tap from a plurality of memory taps in the neural network. Further, the method includes modifying the input data based on the determined inverse function value by dynamically changing a usage of the at least one memory tap from the plurality of memory taps. Further, the method includes compensating the non-linear effect in the output data of the power amplifier.
1 . A method for controlling a non-linear effect of a power amplifier, the method comprising:
training, by an apparatus, a neural network (NN)-based Digital Pre-distorter (DPD) of the apparatus, wherein the NN-based DPD comprises at least one sub-network from a plurality of sub-networks;
controlling, by the apparatus, the non-linear effect of output data of the power amplifier using the trained NN-based DPD;
transmitting, by the apparatus, a linearly amplified signal comprising the non-linear effect of the output data of the power amplifier;
monitoring, by the apparatus, at least one of an error vector magnitude (EVM) parameter associated with the linearly amplified signal or an adjacent channel leakage ratio (ACLR) parameter associated with the linearly amplified signal;
determining, by the apparatus, whether at least one of the EVM parameter and the ACLR parameter satisfies a specified threshold; and
when at least one of the EVM parameter and the ACLR parameter does not satisfy the specified threshold, retraining the trained NN-based DPD.
2 . The method as claimed in claim 1 , wherein the method further comprises:
determining, by the apparatus, whether at least one of the EVM parameter and the ACLR parameter satisfies the specified threshold; and
performing, by the apparatus, at least one of:
retaining at least one NN parameter in the trained NN-based DPD when at least one of the EVM parameter and the ACLR parameter satisfies the specified threshold, and
retraining the trained NN-based DPD to modify at least one NN parameter in the trained NN-based DPD when at least one of the EVM parameter and the ACLR parameter does not satisfy the specified threshold.
3 . The method as claimed in claim 1 , wherein a number of an output node of the sub-network corresponds to a current input data, to at least one previous input data and at least one previous output data.
4 . The method as claimed in claim 1 , wherein the NN-based DPD comprises:
training a fully connected part of the network, wherein the fully connected part captures a non-linear parameter, wherein the non-linear parameter comprises memory information and temperature information; and
training a partially connected part of the plurality of sub-networks based on at least one previous input data and at least one previous output data, wherein the fully connected part and the partially connected part are trained separately, wherein the fully connected part is trained in real time, wherein the fully connected part of the network and the partially connected part of the plurality of sub-networks are trained together in an initial stage, and based on there being further non-linearities in the apparatus, the apparatus is configured to train only the fully connected part while retaining previous parameters for the partially connected part in the real time.
5 . The method as claimed in claim 1 , wherein the NN-based DPD is trained by all sub-networks corresponding to at least one previous input data and at least one previous output data with at least one sub-network parameter, wherein the at least one sub-network parameter comprises an output node, weight and biases.
6 . A method for controlling a non-linear effect of a power amplifier, the method comprising:
acquiring, by an apparatus, an input data of the power amplifier (PA) of the apparatus and an output data of the power amplifier;
determining, by the apparatus, an inverse function using a neural network, wherein the inverse function maps normalized output data of the PA to the input data of the PA, wherein the neural network comprises at least one sub-network for each of at least one memory tap from a plurality of memory taps in the neural network;
modifying, by the apparatus, the input data based on the determined inverse function by dynamically changing a usage of the at least one memory tap from the plurality of memory taps; and
compensating, by the apparatus, the non-linear effect in the output data of the power amplifier.
7 . The method as claimed in claim 6 , wherein the inverse function is learned by reducing an error between a current input of the power amplifier and an estimated input of the power amplifier from a normalized power amplifier output over a period of time.
8 . An apparatus configured to control a non-linear effect of a power amplifier, the apparatus comprising:
a neural network (NN)-based Digital Pre-distorter (DPD), wherein the NN-based DPD comprises at least one sub-network from a plurality of sub-networks; and
a trained NN-based DPD disposed before the power amplifier and configured to control the non-linear effect of output data of the power amplifier,
wherein the power amplifier is configured to transmit a linearly amplified signal comprising the non-linear effect of the output data of the power amplifier, and
wherein the NN-based DPD is configured to:
monitor at least one of an error vector magnitude (EVM) parameter associated with the linearly amplified signal and an adjacent channel leakage ratio (ACLR) parameter associated with the linearly amplified signal,
determine whether at least one of the EVM parameter and the ACLR parameter satisfies a specified threshold, and
when at least one of the EVM parameter and the ACLR parameter does not satisfy the specified threshold, retrain the trained NN-based DPD.
9 . The apparatus as claimed in claim 8 , wherein the trained NN-based DPD is configured to:
determine whether at least one of the EVM parameter and the ACLR parameter satisfies the specified threshold; and
perform at least one of:
retaining at least one NN parameter in the trained NN-based DPD when at least one of the EVM parameter and the ACLR parameter satisfies the specified threshold, and
retraining the trained NN-based DPD to modify at least one NN parameter in the trained NN-based DPD when at least one of the EVM parameter and the ACLR parameter does not satisfy the specified threshold.
10 . The apparatus as claimed in claim 8 , wherein a number of an output node of the sub-network corresponds to a current input data, to at least one previous input data and at least one previous output data, wherein the number of output nodes corresponding to the sub-networks for current and past samples are different.
11 . The apparatus as claimed in claim 8 , wherein the NN-based DPD is configured to be trained by:
training a fully connected part of the network, wherein the fully connected part captures a non-linear parameter, wherein the non-linear parameter comprises a memory information and temperature information; and
training a partially connected part of the plurality of sub-networks based on at least one previous input data and at least one previous output data, wherein the fully connected part and the partially connected part are trained separately, wherein the fully connected part is trained in real time, wherein the fully connected part of the network and the partially connected part of the plurality of sub-networks are trained together in an initial stage, and based on there being further non linearities in the apparatus, the apparatus is configured to train only the fully connected part while retaining previous parameters for the partially connected part in the real time.
12 . The apparatus as claimed in claim 8 , wherein the NN-based DPD is configured to be trained by all sub-networks corresponding to at least one previous input data and at least one previous output data with at least one sub-network parameter, wherein the at least one sub-network parameter comprises an output node, weight and biases.
13 . An apparatus configured to control a non-linear effect of a power amplifier, the apparatus comprising:
a neural network (NN)-based Digital Pre-distorter (DPD), wherein the NN-based DPD comprises at least one sub-network from a plurality of sub-networks, wherein the NN-based DPD is configured to:
acquire an input data of the power amplifier (PA) of the apparatus and an output data of the power amplifier;
determine an inverse function, wherein the inverse function is configured to map normalized output data of the PA to the input data of the PA wherein the NN-based DPD comprises at least one sub-network for each of at least one memory tap from a plurality of memory taps in the neural network;
modifying the input data based on the determined inverse function by dynamically changing a usage of the at least one memory tap from the plurality of memory taps; and
compensating for the non-linear effect in the output data of the power amplifier.
14 . The apparatus as claimed in claim 13 , wherein the inverse function is learned by reducing an error between a current input of the power amplifier and an estimated input of the power amplifier from a normalized power amplifier output over a period of time.