Network energy efficiency is of critical importance to mobile network operators for economic and ecological reasons. The advent of the O-RAN architecture has brought disaggregation and virtualization, and in order to achieve the highest energy savings gains, we need rigorous measurement, analysis, and modeling of energy consumption at both the component and system levels. However, there remains a lack of publicly-available, quantitative data characterizing the behavior of commercial-grade O-RAN systems. In this white paper, we present a detailed energy-efficiency characterization and modeling of a commercial O-RAN system based on comprehensive power and performance measurements, using a network deployment that faithfully replicates a production O-RAN network deployed by a wireless carrier. The results are drawn from an energy test campaign conducted through a joint collaboration between the Open RAN Center for Integration and Deployment (ORCID) Lab Testing and Evaluation (T&E) Project and the Open Networking Foundation / Rutgers WINLAB Energy Efficiency R&D project. The test environment includes an O-RAN system with an AWS-hosted O-CU, a dedicated-server O-DU, and six high-power, multi-band O-RUs. Our results identify the dominant factors influencing power consumption across the O-RAN stack and quantify energy usage variation under different operational and traffic scenarios. These measurements can be used by operators to parameterize power-consumption models, ultimately supporting data-driven energy optimization and more sustainable operation of commercial O-RAN networks.
Alex: Welcome to another episode of ResearchPod. Sam, we've been talking about making mobile networks more efficient—today, what's on the table?
Sam: This white paper examines energy use in a real-world commercial O-RAN system. O-RAN splits a 5G cell tower's work into separate pieces—like the radio hardware that sends signals and software on computers that handles the data processing. This setup lets companies mix gear from different makers and adjust things flexibly, but it makes tracking electricity costs trickier.
Alex: So the main puzzle here is figuring out how much power these split-up systems actually draw?
Sam: Yes, exactly. Mobile network operators build these O-RAN setups for macro sites—big towers covering cities—but without solid data on power draw during changing traffic, they risk surprise bills and higher emissions. The paper fills that gap with detailed measurements from a test setup mimicking a carrier's network, including cloud-based processing units and multiple radio units.
Alex: Right, because more towers and fancy antennas already guzzle energy.
Sam: That's correct. About three-quarters of a network's power goes to those spread-out base stations, and O-RAN's changes add complexity since physical radio parts behave differently from virtual software ones. The study provides the first public quantitative look at this, pinpointing what drives power use across the pieces.
Alex: And without models to predict it?
Sam: Operators fly blind on sustainability. This work offers a power model for the radio units, fit from simple tests, to help forecast and cut waste while keeping service steady.
Alex: So this power model for the radio units—how does it actually work? Walk me through the pieces.
Sam: Picture the radio unit like a house that always needs some baseline electricity just to stay lit—fans running, clocks ticking. Then, for each radio frequency band it uses, like different rooms, there's extra power for idle lights in those rooms plus the cost of appliances that ramp up when sending signals. The model adds those up: fixed house power, plus per-band idle draw, plus signal output divided by how efficiently each band's amplifier turns electricity into radio waves. They fit the numbers from straightforward tests, like running single carriers at different loads.
Alex: Okay, so the amplifiers are the big variable depending on the band? That makes sense for multi-band units handling different frequencies.
Sam: Precisely. Type-A units juggle two bands, one with dual carriers needing beefier amps; Type-B covers three, but tests focused on two active ones. Measurements came from external meters on DC power feeds—ground truth down to about 5 watts—cross-checked against the units' own reports, which matched within 10%. This let them predict total draw across test cases varying transmit power, MIMO layers, and carrier counts.
Alex: And those test cases showed the model holding up?
Sam: Yes, with under 1% error in predictions. As more units and carriers kick in, fixed overheads from processing units spread out, boosting overall efficiency—reaching about 28% better scaling in six-unit setups versus one. The paper stresses this as a snapshot from one commercial system, not a vendor showdown, but a validated way to parameterize from simple idle and load tests.
Alex: Huh. So operators get a tool to forecast without full-scale trials every time.
Sam: Yes. For instance, with no signals on the N70 band alone, power hit 207 watts; N66g alone was 236 watts; both active reached 291 watts. Subtracting those gives 152 watts for the fixed base, 55 watts idle for N70's chain, and 84 watts for N66g's. That higher idle on N66g ties to its beefier setup for dual carriers and higher-rated amplifiers.
Alex: Huh, so the bands aren't equal—some stages just draw more even when waiting. What about when signals ramp up?
Sam: Exactly, and that's where amplifier efficiency comes in—the fraction of input electricity that turns into useful radio output, like how much of your bike pedaling actually moves you forward versus wasted heat. For N70, it ranged from 29 percent at lower output to 39 percent at peak; N66g was lower, up to 32 percent. Dual-band blends those, matching measurements closely.
Alex: Okay, that fits the model's pieces. But the processing side—the O-DU on that server—did it scale the same way with more units?
Sam: Not quite as variable. They measured the full Dell server power, plus estimates for just the O-DU software pods in Kubernetes—a container system like virtual rooms sharing one house computer. Idle baseline stayed steady, with modest bumps: about 10 watts extra for two units, 25 watts for six. Much of the server's 130-watt gap beyond pods is fixed overhead like fans.
Alex: So the extra processing load doesn't spike much, even at full tilt.
Sam: Right—proportional to baseline, it's small, under 10 percent rise. The O-CU on AWS cloud showed even less: data traffic adding just 1 watt. Overall, as radio units scale up, their efficiency gains outweigh these modest central increases.
Alex: That lines up with spreading fixed costs thinner. But does that actually show up in how efficiently the whole system turns power into data delivery?
Sam: Yes, and the paper measures that directly. They define energy efficiency as the amount of data successfully sent—total bits delivered to phones—divided by the total electricity used across all parts, like miles driven per gallon of gas for the network. For their full setup with six radio units pushing maximum data, it hit 655 kilobits per second per watt—a clear step up from 512 with just one unit, because the central processing's steady draw gets shared over way more data flow.
Alex: Okay, so bigger scale pays off by diluting the always-on costs. What happens if traffic isn't maxed out, like real-world partial demand?
Sam: That's a key insight. When they cut data demand to half, throughput halved but total power barely budged—dropping just 2 to 5 percent—since idle draws in radios and processing dominate. Efficiency plunged nearly in half. The logic is straightforward: without features to deeply sleep unused parts, you're still paying full overhead for half the work.
Alex: Huh, so running flat-out is way better than idling half-empty. Does tweaking antennas help there?
Sam: It can, but marginally. Dropping from four antenna streams per direction to two halved data rate and radio output power, trimming unit draw by 12 percent, yet system efficiency still fell 48 percent—idle overheads overwhelm the savings. Higher streams pack more data efficiently when demand justifies it.
Alex: And distance to phones? Does signal weakening change the power side?
Sam: No power shift there. Medium or high path loss—fading signals over distance—lowers data packing density and throughput without cutting consumption, tanking efficiency further. Best results demand close, high-demand conditions.
Alex: So the model's predictions highlight running full tilt across scales to beat those overheads.
Sam: Precisely. The study pinpoints baseline overheads, transmitted radio power, and amplifier efficiency as the main drivers—validated across their test cases from one to six units.
Alex: Right, pulling it all together with that model. But as a snapshot from one commercial setup, what are the edges we should keep in mind?
Sam: Fair point. It's from a single end-to-end system, so not comparing vendors; no energy-saving modes like deep sleep were active; it skips cloud hosting beyond the central unit and core network; and tests used static loads, not real-world traffic swings. That keeps predictions tight for similar setups but calls for broader validation.
Alex: Huh. Ties right into making 5G greener without guessing on bills or emissions.
Sam: Exactly. Energy efficiency matters for sustainable networks, and disaggregation offers chances if backed by data like this—actionable for the O-RAN community worldwide. The models and methods here support ongoing work toward that.
Alex: A clear step for handling those split-up systems practically. Thanks, Sam—that's our look at power modeling in commercial O-RAN. Thanks for listening to ResearchPod.