US 12,688,413 B2 · Methods for reliable over-the-air computation with pulses for distributed learning and with federated edge learning without channel state information

Abstract

An over-the-air computation (AirComp) scheme is proposed for federated edge learning (FEEL) without channel state information (CSI) at the edge devices (EDs) or edge server (ES). The proposed scheme adopts the majority vote (MV) principle and uses pulse-position modulation (PPM) symbols constructed with discrete Fourier transform (DFT)-spread orthogonal frequency division multiplexing (OFDM) (DFT-s-OFDM) as votes from EDs. By taking the delay spread and synchronization errors into account, we show how to eliminate the need for truncated-channel inversion (TCI) at the EDs and detect MV at the ED with a non-coherent detector. The proposed method naturally reduces the peak-to-mean envelope power ratio (PMEPR) of the signal as it inherits the properties of the single-carrier (SC) waveform. An alternative proposed scheme also adopts the majority vote (MV) principle but further defines multiple subcarriers and orthogonal frequency division multiplexing (OFDM) symbols for voting options, which reduces to frequency-shift keying (FSK) over OFDM subcarriers as a special case. Since the votes from EDs are separated on orthogonal resources, the proposed scheme eliminates the need for truncated-channel inversion (TCI) at the EDs and allows the ES to detect MV with a non-coherent detector. We also mitigate the peak-to-mean envelope power ratio (PMEPR) of the synthesized signals by using randomization symbols. Through simulations, we show that the proposed schemes provide high test accuracy in fading channels for both independent and identically distributed (IID) and non-IID data while resulting in lower PMEPR symbols as compared to one-bit broadband digital aggregation (OBDA) with quadrature amplitude modulation (QAM).

Claims36

Dependent claims

An over-the-air computation (AirComp) methodology for federated edge learning (FEEL) without using channel state information (CSI) at a plurality of edge devices (EDs) or at an edge server (ES), comprising:

a distributed machine-learning model to be trained with the update vectors received at an edge server (ES) as transmitted from a plurality of edge devices (EDs);
one or more processors at the plurality of EDs and the ES;
and
one or more non-transitory computer-readable media that store instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, the operations comprising:
at each respective ED, encoding local update vectors into votes by mapping voting options to separate orthogonal physical resources, the votes comprising (1) pulse-position modulation (PPM) symbols synthesized with discrete Fourier transform (DFT)-spread orthogonal frequency division multiplexing (OFDM) (DFT-s-OFDM) or (2) frequency-shift keying (FSK) symbols constructed with orthogonal frequency division multiplexing (OFDM),
transmitting the encoded local update vectors as votes over selected multiple physical resources from each respective of the plurality of edge devices (EDs) via a wireless multiple access channel on common time-frequency resources,
receiving the superposed local updates at the ES as a combined wireless signal,
at the ES, determining the majority vote (MV) for each element of the update vector non-coherently with an energy detector by comparing energy levels across the orthogonal physical resources to resolve the MV without performing channel inversion at the plurality of EDs and without using channel state information (CSI) at the ES, and
inputting the MVs into the machine-learning model to be updated.

6 more elements

An over-the-air computation (AirComp) system for federated edge learning (FEEL) without using channel state information (CSI) at a plurality of edge devices (EDs) or at an edge server (ES), comprising:

a machine-learning model training to process data received at an edge server (ES) as transmitted from a plurality of edge devices (EDs);
one or more processors at the plurality of EDs and the ES;
and
one or more non-transitory computer-readable media that store instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, the operations comprising:
at each respective ED, encoding local update vectors into votes by mapping voting options to separate orthogonal physical resources, the votes comprising (1) pulse-position modulation (PPM) symbols synthesized with discrete Fourier transform (DFT)-spread orthogonal frequency division multiplexing (OFDM) (DFT-s-OFDM) or (2) frequency-shift keying (FSK) symbols constructed with orthogonal frequency division multiplexing (OFDM),
transmitting the encoded local updates as votes over selected multiple physical resources from each respective of the plurality of edge devices (EDs) via a wireless multiple access channel on common time-frequency resources,
receiving the superposed local updates at the ES as a combined wireless signal,
aggregating the local updates at the ES non-coherently including separating votes from the EDs by comparing energy levels across the orthogonal physical resources to resolve a majority vote (MV) without performing channel inversion at the plurality of EDs and without using channel state information (CSI) at the ES, and
inputting the obtained data into the machine-learning model as training data or data to process.

6 more elements

a machine-learning model training to process data received at an edge server (ES) as transmitted from a plurality of edge devices (EDs);
one or more processors at the plurality of EDs and the ES;
and
one or more non-transitory computer-readable media that store instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, the operations comprising:
at each respective ED, encoding local updates into votes by mapping voting options to separate orthogonal physical resources, wherein the votes comprise frequency-shift keying (FSK) symbols constructed with orthogonal frequency division multiplexing (OFDM) for voting options,
transmitting the encoded local updates as votes over the orthogonal physical resources from each respective of the plurality of edge devices (EDs) via a wireless multiple access channel on common time-frequency resources,
receiving the superposed local updates at the ES as a combined wireless signal,
aggregating the local updates at the ES non-coherently including separating votes from the EDs by comparing energy levels across the orthogonal physical resources to resolve a majority vote (MV) principle without performing channel inversion at the plurality of EDs and without using channel state information (CSI) at the ES, and
inputting the obtained data into the machine-learning model as training data or data to process.

6 more elements

Dependencies are inferred from the claim text.

Methods for reliable over-the-air computation with pulses for distributed learning and with federated edge learning without channel state information

Life of this patent

Methods for reliable over-the-air computation with pulses for distributed learning and with federated edge learning without channel state information

Life of this patent

Abstract

Claims36