Wireless devices and systems including examples of compensating power amplifier noise with neural networks or recurrent neural networks
View Patent ↗Examples described herein include methods, devices, and systems which may compensate input data for nonlinear power amplifier noise to generate compensated input data. In compensating the noise, during an uplink transmission time interval (TTI), a switch path is activated to provide amplified input data to a receiver stage including a recurrent neural network (RNN). The RNN may calculate an error representative of the noise based partly on the input signal to be transmitted and a feedback signal to generate filter coefficient data associated with the power amplifier noise. The feedback signal is provided, after processing through the receiver, to the RNN. During an uplink TTI, the amplified input data may also be transmitted as the RF wireless transmission via an RF antenna. During a downlink TTI, the switch path may be deactivated and the receiver stage may receive an additional RF wireless transmission to be processed in the receiver stage.
1. A method, comprising:
amplifying a first transmission signal using a power amplifier;
providing, from an output of the power amplifier and after activating a switch path, the first transmission signal as feedback to a recurrent neural network (RNN); and
mixing, at the RNN, the feedback as input data using a plurality of coefficients to generate filter coefficient data that at least partially compensates nonlinear power amplifier noise in transmission signals, wherein generating the filter coefficient data includes computing, by the RNN, an error signal for reducing a difference between the feedback and a signal to be transmitted, and wherein first processing results are calculated based on the input data and delayed versions of respective outputs of a first plurality of processing units and output data is calculated based on the first processing results and delayed versions of respective outputs of an additional plurality of processing units.
2. The method of claim 1 , further comprising:
activating the switch path based at least in part on receiving a selection signal; and
receiving the selection signal that indicates the switch path is to be activated when a transmission time interval (TTI) of a time-division duplex (TDD) configured radio frame is designated for an uplink transmission.
3. The method of claim 1 , further comprising:
filtering a second transmission signal in accordance with the filter coefficient data, prior to amplifying the second transmission signal using the power amplifier.
4. A method comprising:
amplifying a first transmission signal using a power amplifier;
providing from an output of the power amplifier and after activating a switch path the first transmission signal as feedback to a recurrent neural network (RNN); and
mixing, at the RNN, the feedback as input data using a plurality of coefficients to generate filter coefficient data that at least partially compensates nonlinear power amplifier noise in transmission signals, wherein generating the filter coefficient data includes computing, by the RNN, an error signal for reducing a difference between the feedback and a signal to be transmitted, wherein mixing, at the RNN, the input data comprises:
calculating first processing results based on the input data and delayed versions of respective outputs of a first plurality of processing units with the plurality of coefficients, wherein the first plurality of processing units perform multiplication, accumulation, or both multiplication and accumulation; and
calculating output data based on the first processing results and delayed versions of respective outputs of a respective additional plurality of processing units with an additional plurality of coefficients.
5. The method of claim 4 , further comprising:
providing, from the RNN, the output data as the filter coefficient data to a digital filter that uses the filter coefficient data to at least partially compensate the nonlinear power amplifier noise in the transmission signals.
6. A method of compensating for nonlinear power amplifier noise in transmission signals, comprising:
amplifying a first transmission signal using a power amplifier,
providing feedback from an output of the power amplifier to a recurrent neural network (RNN),
generating, by the RNN, filter coefficient data that compensates for the nonlinear power amplifier noise, wherein generating the filter coefficient data includes computing, by the RNN, an error signal for reducing a difference between the feedback and a signal to be transmitted, wherein the feedback is mixed as input data using a plurality of coefficients to generate the filter coefficient data, and wherein processing results are calculated based on the input data and delayed versions of respective outputs of a plurality of processing units, and
applying the filter coefficient data to a second transmission signal prior to amplifying the second transmission signal using the power amplifier.
7. The method of claim 6 , further comprising activating a switch path before providing the feedback from the output of the power amplifier to the RNN.
8. The method of claim 6 , wherein generating the filter coefficient data comprises using a plurality of coefficients and an additional plurality of coefficients by the RNN.
9. The method of claim 6 , wherein generating, by the RNN, the filter coefficient data comprises mixing the feedback as input data with a plurality of coefficients.
10. The method of claim 6 , further comprising providing the filter coefficient data from the RNN to a digital filter that uses the filter coefficient data to compensate the nonlinear power amplifier noise in the transmission signals.
11. The method of claim 6 , wherein the power amplifier comprises a nonlinear power amplifier.
12. The method of claim 6 , wherein the filter coefficient data compensates for the nonlinear power amplifier noise in a digital pre-distortion model.
13. A method comprising:
amplifying a first transmission signal using a power amplifier;
providing feedback from an output of the power amplifier to a recurrent neural network (RNN);
generating, by the RNN, filter coefficient data that compensates for nonlinear power amplifier noise, wherein generating the filter coefficient data includes computing, by the RNN, an error signal for reducing a difference between the feedback and a signal to be transmitted, wherein generating, by the RNN, the filter coefficient data comprises mixing the feedback as input data with a plurality of coefficients;
applying the filter coefficient data to a second transmission signal prior to amplifying the second transmission signal using the power amplifier; and
calculating, by the RNN, processing results based on the input data and delayed versions of respective outputs of a first plurality of processing units with the plurality of coefficients, wherein the first plurality of processing units perform multiplication, accumulation, or both multiplication and accumulation.
14. The method of claim 13 , further comprising calculating, by the RNN, output data based on the processing results and delayed versions of respective outputs of a respective additional plurality of MAC units with the additional plurality of coefficients.
15. A method of improving uplink transmission in a time-division duplex (TDD) configured radio frame, comprising:
amplifying a first transmission signal using a power amplifier,
activating a switch path based on a selection signal indicating an uplink transmission time interval (TTI) in the TDD configured radio frame,
providing feedback from an output of the power amplifier to a recurrent neural network (RNN), and
generating, by the RNN, filter coefficient data that compensates for nonlinear power amplifier noise, wherein generating the filter coefficient data includes computing, by the RNN, an error signal for reducing a difference between the feedback and a signal to be transmitted, wherein the feedback is mixed as input data using a plurality of coefficients to generate the filter coefficient data, and wherein processing results are calculated based on the input data and delayed versions of respective outputs of a plurality of processing units.
16. The method of claim 15 , wherein the generating comprises mixing the feedback as input data using at least a first layer of processing units and a second layer of processing units, wherein the first layer of processing units perform multiplication, accumulation, or both, and wherein the second layer of processing units comprises memory look-up units (MLUs).
17. The method of claim 15 , wherein the filter coefficient data is used to reduce error in a digital pre-distortion model.
18. The method of claim 17 , wherein the error reduction is based on reducing a difference between a signal to be transmitted and the feedback.
19. The method of claim 18 , wherein the signal to be transmitted is based at least in part on a nonlinear order of the power amplifier noise to be compensated.
20. The method of claim 15 , wherein the activating the switch path comprises activating the switch during the same time or during a portion of the same time as transmitting the first transmission signal to an antenna.