The 6.8-Watt Antenna That Could Make Smart Roads Affordable
The Communication Gap in Our Roads
Somewhere in Beijing, a small robot vehicle is rolling along a predefined path. At regular intervals, a roadside antenna array—technically a reconfigurable intelligent surface, or RIS—pings the vehicle with radio waves, and the echoes come back carrying two kinds of information simultaneously: where the vehicle is, and whether it's receiving data cleanly. The position estimate is accurate to 11 centimeters. The data arrives with 42 percent less distortion than it would without the RIS. And the whole roadside system—the antenna, the signal processing, the computing—draws less power than a dim desk lamp.
That 6.8 watts is the number that should get attention. Because for years, the wireless industry's dream of combining radar-like sensing with high-speed communication has run headlong into a physical problem: you need a lot of antennas to make it work well, and lots of antennas mean lots of phase shifters, and lots of phase shifters mean lots of power. Building the large-aperture antenna arrays that next-generation vehicular networks require—arrays capable of tracking cars with centimeter precision while streaming video to dashboards—was becoming an energy and cost nightmare.
A team from Peking University has just demonstrated something that suggests the nightmare may be ending. Their prototype, described in a new paper published on arXiv, represents the first real-world implementation of a technology that could solve both problems at once: a reconfigurable intelligent surface, or RIS, integrated with what engineers call ISAC—integrated sensing and communication. And the results, measured in an actual outdoor test rather than a simulation, suggest that the technology is not merely promising but practically deployable (Yang et al., 2026).
The Infrastructure of Connected Roads
Before diving into what the Peking team built, it helps to understand why the problem exists in the first place. Future roads—the ones that will eventually host autonomous vehicles, cooperative collision avoidance, and real-time traffic optimization—need infrastructure that can do two things simultaneously: communicate with cars and sense their positions.
Communication is straightforward enough: the roadside unit, or RSU, sends data to vehicles, and the vehicles send data back. But sensing—the ability to track where vehicles are, how fast they're moving, what trajectory they're on—requires something more. It requires the roadside infrastructure to bounce radio waves off vehicles and interpret the echoes. The physics of radio detection and ranging, or radar, tells us that you get better resolution—meaning you can distinguish smaller objects and measure positions more precisely—when you have a larger antenna aperture. More antenna elements, carefully coordinated, means sharper beams and more accurate tracking.
The traditional way to build a large-aperture antenna array is with phased array technology. You arrange hundreds or thousands of antenna elements in a grid, connect each one to a phase shifter, and then carefully adjust the phase of the signal at each element so that the waves constructively interfere in the direction you want to point. It's powerful technology, and it's mature. It's also power-hungry. Each phase shifter draws current, and when you're dealing with the kinds of arrays that future vehicular networks will need—arrays with enough aperture to track vehicles across hundreds of meters—you're talking about power consumption that makes widespread deployment impractical.
The alternative the Peking team chose is a technology that has generated enormous excitement in the wireless research community over the past several years: the reconfigurable intelligent surface. An RIS is, at its core, a sheet of material embedded with a large number of small, inexpensive reflecting elements. Each element can be programmed to change how it reflects incoming radio waves. By adjusting the phase of the reflection at each element according to a carefully calculated pattern, the entire surface can act like a lens, directing incoming radio energy into a concentrated beam pointing wherever you need it.
The key advantage is what's NOT there: phase shifters drawing power at each element. Instead, the RIS elements use diode-based tunable components that consume trivial amounts of power. The Peking team's surface contains 1,600 individual elements arranged in a 40-by-40 grid. It operates at 5.5 GHz—firmly in the sub-6 GHz band used by existing cellular networks, which means it's compatible with hardware and spectrum allocations that already exist. And the total power consumption for the entire surface is 6.8 watts.
Building the Prototype
The system the researchers built has two main components: a roadside node and a vehicle terminal.
The roadside node is where the intelligence lives. It centers on a USRP X310—a software-defined radio platform that can generate and receive arbitrary wireless signals under software control. The USRP connects to a power amplifier that boosts the transmit power to 35 decibel-milliwatts (dBm), then to a feed antenna that illuminates the RIS. The RIS, positioned a quarter-meter from the feed antenna at a 30-degree elevation angle, reflects that incoming energy toward the vehicle. A separate receive antenna on the roadside node captures the echoes that bounce back from the vehicle, and the USRP digitizes these for processing.
The vehicle terminal is simpler: it rides on a small robotic vehicle that moves along a predefined path, carrying a receive antenna and its own USRP to demodulate the incoming communication signals. The robot starts 3 meters from the RIS at a 45-degree azimuth angle—roughly where a car might approach an intersection—then moves along a curved trajectory and stops about 2.5 meters away at a 10-degree angle.
The signal format the team chose is worth noting. They used orthogonal frequency division multiplexing, or OFDM—the same digital modulation scheme used in 4G LTE, 5G, and Wi-Fi—with 1,024 subcarriers spread across a 100-megahertz bandwidth. For the communication data, they used 16-QAM modulation, which packs four bits into each transmitted symbol. This isn't the most sophisticated modulation available—later vehicular systems will likely use higher-order schemes to achieve faster data rates—but it's well-calibrated for a first demonstration: robust enough to work reliably in a challenging outdoor propagation environment while being complex enough to reveal real differences between the RIS-assisted system and a baseline.
The overall protocol works in two stages. In the initialization phase, the roadside node transmits an ISAC signal—that is, a signal designed to serve double duty as both a communication transmission and a radar pulse—and cycles the RIS through a series of beam patterns, scanning across the entire region of interest. For each beam direction, the node processes the returned echo using OFDM radar processing techniques to build up a range-angle heatmap: a two-dimensional picture showing where reflectors are located in distance and direction. A detection algorithm called CFAR—constant false alarm rate—identifies the strongest returns, extracts the vehicle's initial position, and then configures the RIS to focus its beam directly at that position. This establishes a communication link with enough signal strength to be useful.
Then the system transitions to the adaptive beam-tracking stage. Over multiple rounds, the RIS performs finer-grained scans, but only within a small angular window around where it last estimated the vehicle to be. This dramatically reduces the search overhead compared to scanning the entire region each time. The system also applies an extended Kalman filter—essentially a mathematical technique for estimating the true position of a moving object by fusing a sequence of noisy measurements with a model of how objects typically move—to refine the position estimates and predict where the vehicle will be next. This predicted position determines the scanning window for the following round.
The result is a closed-loop system: the roadside unit tracks the vehicle, predicts where it's going, points its beam there, maintains a high-quality communication link, and repeats this process every few milliseconds as the vehicle moves.
The Numbers
The experiment ran. The robot moved. And the measurements came back showing clear, substantial improvements.
Let's start with the communication performance, because this is where the 41.9 percent figure comes from. The metric the researchers used is called error vector magnitude, or EVM—a measure of how much the received constellation points deviate from their ideal positions. A constellation is the pattern that the received symbols make when plotted on a complex plane; 16-QAM produces a grid of 16 points, and noise, interference, and distortion cause those points to smear. The more they smear, the higher the EVM, and the more likely it is that bits will be received in error.
Error Vector Magnitude: Without RIS vs. With RIS
Error vector magnitude comparison between baseline and RIS-assisted system, showing a 41.9% reduction in signal distortion.
| Label | Value |
|---|---|
| Without RIS | 17.35 |
| With RIS | 10.08 |
With the RIS beam optimized to point at the vehicle, the EVM dropped from 17.35 percent to 10.08 percent. That's a 41.9 percent reduction in distortion. But the researchers also measured something more concrete: the block error rate, or BLER. Instead of looking at how much individual symbols are distorted, BLER counts how many complete blocks of data arrive with any errors at all. Here the improvement was even more dramatic: from 5.14 percent without the RIS to 0.73 percent with it. That's a factor of seven reduction in error rate.
The clear constellation display in Figure 3(a) tells the story visually: the 16 points of the 16-QAM constellation are crisp and well-separated, indicating that the vehicle can receive video data with minimal errors.
Block Error Rate: Without RIS vs. With RIS
Block error rate comparison showing dramatic reduction in data transmission errors when using RIS-assisted beamforming.
| Label | Value |
|---|---|
| Without RIS | 5.14 |
| With RIS | 0.73 |
But the more surprising result is the localization accuracy. The average error in estimating the vehicle's position was 0.11 meters—eleven centimeters. To put that in perspective: that's roughly the width of a human hand. For a system that's bouncing radio waves off a moving robot to determine its location, achieving centimeter-scale accuracy without any specialized radar hardware is a significant achievement. The trajectory plot in Figure 3(b) shows the estimated path (in orange) closely tracking the actual path (in blue), with the red dot indicating the vehicle's current estimated position.
Average Vehicle Localization Error
Average vehicle position error achieved with the RIS-ISAC prototype in real-world outdoor testing.
| Label | Value |
|---|---|
| Localization Accuracy | 0.11 |
Why This Changes Things
Here is where the significance becomes clear. The wireless industry has been talking about integrated sensing and communication—ISAC—for years as a key enabler of intelligent transportation systems. The idea is seductive: instead of deploying separate systems for communication and radar, you use a single signal to do both. But there has always been a practical barrier: the antenna arrays needed for high-performance sensing require either huge numbers of phase shifters (expensive and power-hungry) or massive numbers of antenna elements (also expensive, and bulky).
Reconfigurable intelligent surfaces sidestep this problem. Because the RIS doesn't require active electronics at each element—it just reflects incoming waves with programmed phase shifts—the cost and power consumption scale very differently than with a phased array. A 40-by-40 element RIS consumes 6.8 watts. A phased array with the same number of elements would consume orders of magnitude more power, simply because each phase shifter draws current and each phase shifter needs driver electronics and cooling.
This has implications that go beyond the technical performance metrics. Consider what it would take to deploy this technology across a city. A typical urban intersection might need two or three roadside units to provide adequate coverage. If each unit requires a large-aperture antenna array to track vehicles with centimeter precision and maintain high-data-rate communication links, the power and cooling requirements become a significant barrier to deployment. But if those arrays can be implemented with RIS technology at power levels that are essentially negligible—6.8 watts is less than a typical household light bulb—then the economics change entirely.
The sub-6 GHz operating frequency is also a strategic choice. Earlier generations of high-performance wireless sensing systems have often focused on millimeter-wave frequencies—28 GHz, 39 GHz, even higher—because the shorter wavelengths allow for smaller antenna elements and more elements packed into a given area. But millimeter-wave signals have a major weakness: they get absorbed by rain, fog, and even foliage, and they struggle to penetrate walls and other obstacles. Sub-6 GHz signals, by contrast, propagate readily in outdoor environments, penetrate buildings reasonably well, and are already allocated and licensed for cellular use. By demonstrating their system at 5.5 GHz, the Peking team shows a path toward deployment that leverages existing spectrum and existing infrastructure rather than requiring entirely new allocations.
There's also the question of coexistence with existing systems. A vehicular network that operates at 5.5 GHz with 100 MHz of bandwidth can share spectrum with existing cellular networks, using the same frequencies that carriers already pay to license. The ISAC signals don't need to be isolated from Wi-Fi or cellular traffic in the way that dedicated radar systems would need to be—they can be designed to coexist, with appropriate waveforms and protocols, within the existing spectral landscape.
The Practical Path Forward
It's worth being clear about what this demonstration shows and what it doesn't. This is a proof-of-concept prototype operating in a controlled outdoor environment. The vehicle moved along a predefined trajectory in a relatively small area—roughly 3 meters by a few meters—and the system was calibrated for those specific conditions. A real deployment would need to handle much larger coverage areas, many simultaneous vehicles, urban propagation environments with multipath reflections from buildings and other obstacles, and the variability of weather and seasons.
The researchers acknowledge several limitations. The adaptive beam-tracking stage relies on the extended Kalman filter to predict the vehicle's trajectory between measurements. This works well when the vehicle is moving predictably along a known path. It would be more challenging to track a vehicle that's accelerating, braking, or making sudden lane changes. And while the localization accuracy of 0.11 meters is impressive, some applications—fully autonomous driving in particular—may eventually require accuracy at the centimeter level, which would demand even more sophisticated signal processing or higher-resolution sensing modalities.
The system also currently uses a feed antenna to illuminate the RIS from a single point. In larger deployments, it might be desirable to use multiple feed points or to design the RIS with different geometries. And while 6.8 watts is remarkably low for a large-aperture antenna array, the power consumption of the entire roadside unit—including the USRP, the host computer, and the signal processing—isn't specified. A complete system would need to account for all of these components.
But these are engineering challenges, not fundamental barriers. The core result stands: an RIS-based ISAC system can simultaneously track a vehicle with 11-centimeter accuracy and reduce communication errors by 42 percent, all while consuming less power than a typical Wi-Fi router.
System Power Consumption Breakdown
Power consumption breakdown showing the remarkable efficiency of RIS technology compared to conventional phased arrays.
| Label | Value |
|---|---|
| RIS Power Consumption | 6.8 |
| Feed Antenna + PA (est.) | 20 |
| USRP Processing (est.) | 30 |
What Comes Next
The demonstration opens several research directions. On the system design side, there's work to be done on scaling: how would the protocol handle dozens of vehicles simultaneously? How would the beamforming strategies adapt when vehicles are at very different distances or moving in conflicting directions? The current system appears to focus on a single vehicle; a commercial deployment would need to multiplex beams across multiple users.
On the signal processing side, the extended Kalman filter works well for predictable motion, but more sophisticated tracking algorithms may be needed for adversarial or unexpected vehicle behaviors. Machine learning techniques have shown promise in other radar and sensing applications, and it would be natural to explore whether neural networks can learn to predict vehicle trajectories better than linear filters in complex scenarios.
The spectral coexistence question also deserves more attention. The 5.5 GHz band is shared with existing cellular and Wi-Fi systems. An ISAC system that transmits continuously to track vehicles and deliver data will inevitably generate some interference to nearby receivers. How much interference? How far does it propagate? What waveforms and power levels keep the interference acceptably low? These are regulatory and engineering questions that will need answers before widespread deployment.
There's also the question of outdoor-to-indoor coverage. The current demo operates in an open area, but vehicular networks may need to maintain communication links as vehicles pass under overpasses, through tunnels, or into parking structures. Sub-6 GHz signals propagate better through obstacles than millimeter-wave signals, but there's still significant attenuation. An RIS positioned at the entrance to a tunnel could potentially extend coverage and maintain tracking continuity, but this would require careful system design.
Finally, there's the economic question. The researchers built a custom RIS with 1,600 elements. For widespread deployment, these surfaces would need to be manufactured at scale, which would require standardization, testing, and certification. The cost per element would need to come down substantially from what a research prototype costs today. But the semiconductor industry has an excellent track record of reducing costs as volumes increase, and if the demand materializes, there's no obvious reason why RIS elements couldn't follow the same trajectory that Wi-Fi chips, cellular modems, and other wireless components have followed over the past three decades.
A Signal Worth Tracking
The roads of the future will be instrumented in ways that seem almost science-fictional today. Cars will communicate with infrastructure and with each other, sharing sensor data and coordinating maneuvers in real time. Traffic lights will adapt to actual demand rather than timers. Emergency vehicles will be able to preempt traffic, clearing a path before they're even visible. Pedestrians and cyclists will be detected and protected by sensing systems that never look away or get tired.
But all of this requires infrastructure that can both see and speak—and that infrastructure needs to be affordable enough to deploy, and efficient enough to operate at scale. The demonstration from Peking University doesn't solve every problem. But it demonstrates, in a real-world outdoor test, that reconfigurable intelligent surfaces can make integrated sensing and communication practical: power-efficient enough to deploy broadly, accurate enough to be useful, and robust enough to maintain communication quality even as vehicles move through complex environments.
Eleven centimeters of localization error. A 42 percent reduction in communication distortion. Six-point-eight watts of power consumption. These numbers aren't just technical achievements—they're proof points that the technology is maturing from theory to practice. The next time you see a small robot rolling along a predefined path while a roadside antenna tracks its every move, it might be worth remembering: that's a glimpse of the roads we're heading toward.
References
Yang, S., Wang, Y., Zeng, S., Zhang, H., & Di, B. (2026). RIS-Enabled Energy-Efficient ISAC for Vehicular Applications. arXiv preprint. Peking University.