PaperlayerView exact revision on arXiv

Source-linked AI summary

Reducing 6G Energy Consumption: TDD-Aligned Drain Voltage Modulation for Power Amplifiers

Meizhen Zhuo, Gunter Wolff, Stephan ten Brink, Silvio Mandelli

arXiv:2610.32889v1eess.SP

TL;DR

At partial load, fixed-voltage power amplifiers waste supply headroom, while tracking drain voltage to the signal requires a fast, costly supply path. TDD-DVM selects voltage once per TDD period and combines with POLITE link adaptation; at equal delivered throughput, it reduces power while maintaining the full-voltage packet failure ratio.

  • Problem

    At output-power back-off, fixed drain voltage wastes supply headroom as excess DC power, while adapting voltage to the instantaneous signal requires a fast, wideband supply path.

    Source: §I · p. 1

  • Method

    TDD-DVM selects drain voltage from resource load once per TDD period during the uplink, with POLITE lowering transmit power at partial load so a lower voltage can suffice.

    Source: §I · pp. 1–2

  • Results

    At equal delivered throughput, TDD-DVM with POLITE reduces consumed PA power by up to 35% versus full drain voltage under BLA and up to 7% versus full drain voltage under POLITE, with a packet failure ratio indistinguishable from full voltage under POLITE.

    Source: §V · p. 6

  • Takeaways & Limitations

    The scheme achieves power savings at a fraction of the switching cost while capturing roughly half of the Genie-Aided gain.

    Source: §V · p. 6

    The remaining gap to the Genie-Aided bound stems from the resource-load metric's inability to identify the few slots that require full drain voltage.

    Source: §V · p. 6

Abstract

from arXiv · show

The Power Amplifier (PA) of a base station operates inefficiently at partial load, where back-off from the peak output power wastes supply headroom. Adapting the PA drain voltage to the instantaneous signal recovers this loss but requires frequent switching, increasing hardware cost. We propose TDD-Aligned Drain Voltage Modulation (TDD-DVM), a drain voltage modulation scheme that instead selects the drain voltage level once per Time-Division Duplex (TDD) period based on the downlink resource load and switches the voltage during the uplink window. This is enabled by the authors' previous work on POLITE link adaptation, which spreads the transmit power over resources left unused at partial load and moves the PA into a regime where a lower drain voltage suffices. In a 3GPP Urban Macro system-level simulation across traffic loads, we compare TDD-DVM against baseline approaches under conventional and POLITE link adaptation. At equal delivered throughput, TDD-DVM with POLITE keeps the packet failure ratio of the full drain voltage while consuming up to 35% less power than it under conventional link adaptation and up to 7% less than it under POLITE, roughly half of the gain of an idealized reference.

I. INTRODUCTION

Variable traffic leaves base stations over-provisioned, while fixed-voltage power amplifiers lose efficiency at output-power back-off. TDD-DVM selects drain voltage once per TDD period, using POLITE link adaptation to make lower voltage levels sufficient at partial load.

I. INTRODUCTION

  • Over 70% of cellular network power consumption comes from base stations, and their power amplifiers account for a substantial share of dissipation.Traffic varies over time and location, leaving base stations operating below peak capacity for much of the time.Source: §I · p. 1
  • At output-power back-off, fixed drain voltage wastes supply headroom as excess DC power, while lower voltage can improve efficiency but high output still needs headroom for linearity.Signal-following supply modulation can recover back-off efficiency, but Envelope Tracking requires a fast, wideband path and linearization.Source: §I · p. 1
  • TDD-DVM selects drain voltage from current resource load and updates it once per TDD period during the uplink, allowing a slow ramp without signal distortion.The uplink settling interval is much longer than signal-driven tracking timescales, so a slower, cheaper supply path can suffice.Source: §I · pp. 1–2
  • POLITE spreads transmission over otherwise unused resources at lower power under partial load, moving the power amplifier into a regime where lower drain voltage suffices.The paper incorporates POLITE into TDD-DVM and evaluates consumed power, delivered throughput, and packet failure ratio in system-level simulations.Source: §I · pp. 1–2

II. SYSTEM MODEL

The system model represents a multi-cell TDD network with link adaptation determining per-resource transmit power and the resulting power-amplifier consumption. It models traffic, resource allocation, link quality, and delivery against deadlines.

A. TDD Frame Structure

B. Traffic, Resource, and Link Adaptation Model

  • Per-resource transmit power, channel gains, interference, and noise determine SINR; link adaptation selects an MCS against a target BLER, and proportional-fair scheduling distributes resources.The achievable spectral efficiency is smoothed over time using an instantaneous estimate and a smoothing factor.Source: B. Traffic, Resource, and Link… · p. 2
  • Performance is measured by average delivered throughput and packet failure ratio, with delivery assessed against each user's deadline.The failure ratio is the fraction of offered cell traffic not delivered within the deadline.Source: B. Traffic, Resource, and Link… · p. 2

1) Baseline Link Adaptation (BLA):

The baseline link adaptation scheme maximizes spectral efficiency by transmitting at maximum power with the highest-rate MCS that meets the target BLER.

2) Baseline Link Adaptation (BLA):

2) POLITE Link Adaptation:

POLITE adapts transmissions to partial load by spreading data across more resources at lower per-resource power. The power-amplifier model represents discrete drain-voltage levels and their efficiency-dependent consumption.

2) POLITE Link Adaptation:

  • When resources are available at partial load, POLITE spreads data over more resources at lower per-resource power while preserving reliability.Its delay-aware variant also uses the traffic delay budget when deciding how much to spread.Source: 2) POLITE Link Adaptation · p. 2
  • POLITE lowers PA output power at partial load, so a lower drain voltage suffices more often and creates an opportunity for TDD-DVM savings.It does so by choosing a more robust MCS and reducing transmit power to the minimum that still meets the target BLER.Source: 2) POLITE Link Adaptation · p. 2

C. PA Power Consumption Model

  • The PA has discrete drain-voltage levels, with higher levels supporting larger maximum RF output power; its efficiency depends on voltage level and output power.The PA power model applies a single drain-voltage level across the cell's antennas, each of which has its own PA.Source: C. PA Power Consumption Model · p. 3

D. Problem Formulation

The formulation minimizes long-term average downlink PA power while serving offered traffic, with drain-voltage levels held constant within each TDD period and constrained to support requested power.

D. Problem Formulation

  • The objective is to minimize long-term average downlink PA power while still serving the offered traffic.The selected drain-voltage levels must support the scheduler's requested power.Source: D. Problem Formulation · p. 3
  • Drain-voltage level k(t) is constrained to remain constant within each TDD period.Source: D. Problem Formulation · p. 3
  • Every feasible level selection serves offered traffic as the full drain voltage does, while a permanently lower level may be infeasible when requested power exceeds its pmax.Source: D. Problem Formulation · p. 3

III. DRAIN VOLTAGE SELECTION SCHEMES

The section presents reference drain-voltage selection schemes, including a fixed-voltage baseline and per-slot modulation.

III. DRAIN VOLTAGE SELECTION SCHEMES

  • The section introduces reference schemes before presenting the proposed TDD-DVM.Source: §III · p. 3

1) Fixed Drain Voltage:

  • The fixed-voltage reference holds drain voltage constant and postpones traffic that cannot fit within the output-power limit.The scheduler can allocate resources only while output power stays within pmax(k).Source: 1) Fixed Drain Voltage · p. 3

2) Per-Slot DVM (PS-DVM):

  • Per-Slot DVM selects the lowest-power voltage level able to support each downlink slot's output power.Following instantaneous transmit power requires per-slot switching, adding hardware cost.Source: 2) Per-Slot DVM (PS-DVM) · p. 3

3) Genie-Aided:

The Genie-Aided scheme provides a bound for per-period selection using perfect knowledge of upcoming downlink transmissions. Proposed TDD-DVM instead selects a level from resource load and updates the voltage during uplink.

3) Genie-Aided:

  • The Genie-Aided scheme uses the upcoming period's peak output power to select its most energy-efficient supporting level, but requires unavailable perfect advance knowledge.It serves as a bound for per-period selection and cannot outperform per-slot PS-DVM.Source: 3) Genie-Aided · p. 3

B. Proposed Load-Based TDD-DVM With POLITE

  • TDD-DVM ramps drain voltage during uplink, then holds the selected level throughout the following downlink period.The transition finishes before downlink begins; the scheme relies on DU signaling to align the ramp with TDD periods.Source: B. Proposed Load-Based TDD-DVM With… · p. 3, Figure 1 · p. 3
  • The DU selects a target level from resource load, which estimates resources needed to clear buffered traffic relative to those available in a TDD period.A hysteresis rule lowers the level by one step below θlow, raises it by one above θhigh, and otherwise leaves it unchanged.Source: B. Proposed Load-Based TDD-DVM With… · pp. 3–4
  • The algorithm uses per-period TDD-DVM as an outer loop and per-slot POLITE link adaptation as an inner loop for MCS and transmit-power settings.Per-slot consumed power is averaged over the simulation.Source: B. Proposed Load-Based TDD-DVM With… · p. 4

C. Bayesian Optimization (BO)

Bayesian Optimization tunes load-dependent thresholds to balance consumed power against delivery, using a packet-failure constraint referenced to full drain voltage.

C. Bayesian Optimization (BO)

  • The θlow and θhigh thresholds trade energy saving against the risk of choosing a drain voltage too low to serve traffic.Source: C. Bayesian Optimization (BO) · p. 4
  • Bayesian Optimization with Gaussian Processes iteratively evaluates threshold pairs, balancing exploration of uncertain regions with exploitation of promising configurations.Source: C. Bayesian Optimization (BO) · p. 4
  • The thresholds are tuned to minimize consumed power subject to a delivery constraint, with packet failure no worse than full drain voltage at the same traffic load.The resulting threshold pairs are fitted as a traffic-dependent regression curve, and prevailing traffic is estimated from arriving bit rate.Source: C. Bayesian Optimization (BO) · p. 4

IV. NUMERICAL EVALUATION

The evaluation uses a 3GPP 3D Urban Macro system-level simulator and models PA power using measured characteristics of two drain-voltage levels. The setup also compares their measured efficiency across output-power back-off.

A. Evaluation Setup

  • The system-level evaluation follows the 3GPP 3D Urban Macro model, with results averaged over cells in a wrap-around multi-site network.Cells use closed-loop spatial multiplexing, FTP3 traffic, and proportional-fair scheduling.Source: A. Evaluation Setup · p. 4
  • The PA power model uses two measured drain-voltage levels, 80%DV and 100%DV, with Pbase = 15.3% and Pidle = 8.9% for both levels.The measured characteristics report RF output and DC power relative to the 100%DV peak output and consumption, respectively.Source: A. Evaluation Setup · p. 5, Table II · p. 5
  • Fig. 2 plots measured PA efficiency for both drain-voltage levels against output-power back-off from the 100%DV peak.Source: Figure 2 · p. 5

B. Energy Consumption

TDD-DVM combined with POLITE reduces PA power while maintaining equal delivered throughput, with the largest saving measured against Fixed-100%DV under BLA. Its gain is about half the idealized bound at the highest load, at a fraction of the switching cost.

B. Energy Consumption

  • TDD-DVM with POLITE reduces consumed power by up to 7% at equal delivered throughput versus Fixed-100%DV with POLITE, while PS-DVM and the Genie-Aided bound save about 9%.TDD-DVM captures roughly half the ideal gain at the highest load and most of it at moderate loads, at a fraction of the switching cost.Source: B. Energy Consumption · p. 5
  • Fixed-100%DV wastes supply headroom at every load, whereas Fixed-80%DV consumes least but its curve ends near 2.8 Mbps as its power budget limits resource allocation.Source: B. Energy Consumption · p. 5
  • Up to 35% less consumed power than Fixed-100%DV under BLA at equal delivered throughput comes from combining POLITE with TDD-DVM; drain-voltage adaptation contributes about 7%.Fig. 3 plots consumed PA power against average delivered throughput per UE.Source: B. Energy Consumption · p. 5, Figure 3 · p. 5

C. Packet Failure Ratio

TDD-DVM selects drain voltage once per TDD period, with voltage changes during the uplink window. With POLITE, it reduces PA power at equal delivered throughput while maintaining the full-drain-voltage packet failure ratio.

C. Packet Failure Ratio

  • Under POLITE, TDD-DVM's packet failure ratio is indistinguishable from Fixed-100%DV across loads, while Fixed-80%DV reaches 0.22 at the highest load.The same behavior holds under BLA, where both curves lie lower because maximum-power transmission clears queued traffic faster.Source: C. Packet Failure Ratio · p. 5, Figure 4 · p. 6

D. Analysis of Drain Voltage Selection

  • The Genie-Aided bound stays at 80%DV over most of the load range and uses 100%DV in about 16% of slots at highest load, with PS-DVM close by.The figure plots the fraction of downlink slots at 100%DV against offered traffic load per cell.Source: D. Analysis of Drain Voltage Selection · p. 5, Figure 5 · p. 6
  • At the highest load, TDD-DVM selects 100%DV in about 47% of slots under POLITE; whole-period resource load cannot identify the few slots requiring full voltage.Under BLA, TDD-DVM selects 100%DV even more often, while POLITE provides lower-power operation that leaves room for voltage savings.Source: D. Analysis of Drain Voltage Selection · p. 6

V. CONCLUSION

  • TDD-DVM updates drain voltage once per TDD period and ramps it during uplink, allowing a slow ramp without signal distortion and reducing switching hardware cost.Voltage is selected based on resource load before the next downlink transmission opportunity.Source: §V · p. 6
  • At equal delivered throughput, TDD-DVM with POLITE reduces consumed PA power by up to 35% versus full drain voltage under BLA and up to 7% versus it under POLITE.It captures roughly half the Genie-Aided gain without increasing packet failure ratio.Source: §V · p. 6
Loading 2610.32889v1…