When a truck passes between a roadside millimeter-wave (mmWave) base station and a car, the direct radio link can disappear almost instantly. Highway studies attribute sum-rate drops of 40–67% to these events. The usual fix is to bounce the signal off a reconfigurable intelligent surface (RIS), a wall-mounted panel of programmable reflecting elements, so that it travels around the obstacle. That detour has two costs. The reflected path suffers "double fading": its loss scales with the product of the two hop distances rather than their sum. And the channel of the moving car through the surface goes stale within milliseconds. Active surfaces, which embed small amplifiers in each element, recover much of the first cost but add amplified noise. Rate-splitting multiple access (RSMA) is a transmission scheme in which part of every user's data rides in a shared "common" message that all users decode and subtract before decoding their own "private" message. It is known to tolerate stale channel knowledge better than conventional beamforming.
Our proposal starts from the observation that on a highway the timing of blockage is predictable from vehicle sensors and road geometry, and so is the resulting degradation in channel knowledge. We claim that the right share of data to put in the common message can therefore be planned ahead from these large-scale predictions. The share should rise during the window when a user rides the aged, noise-amplified surface detour and fall back afterwards. A plan computed once at the start of a blockage episode should capture most of what per-slot re-optimization with fresh channels would achieve.
Core Claim
Prior work treats the three pieces separately. One line pre-assigns passive RIS units to predicted service regions before a truck arrives (Dynamic Blockage Pre-Avoidance using Reconfigurable Intelligent Surfaces). It uses single-user scheduling and recovers only half of the lost throughput, because the passive detour is double-faded. A second line optimizes active-RIS RSMA (Sum-Rate Maximization for Active RIS-Aided Downlink RSMA System; Active STARS-Assisted Rate-Splitting Multiple-Access Networks) in static settings with perfect or fixed channel knowledge and no blockage. A third line attacks channel aging with predictor antennas on lead vehicles (High-Rate Uninterrupted Internet-of-Vehicle Communications in Highways) but has no surface and no rate splitting. The obvious combination is "active RIS plus RSMA under blockage," and we do not think it is interesting by itself, since RSMA's robustness to imperfect channel knowledge is already established (A Primer on Rate-Splitting Multiple Access).
What is new is the structure we exploit. On a surface detour, the base-station-to-surface hop $G$ is between fixed infrastructure and can be estimated accurately and rarely. Only the surface-to-vehicle hop $v_k$ of user $k$ ages. Under the Jakes model of Doppler fading, the error variance after a delay $\tau$ since the last estimate is
$$\sigma_k^2(\tau) = \big(1 - J_0^2(2\pi f_D \tau)\big)\,\beta_k ,$$
where $J_0$ is the zeroth-order Bessel function, $f_D$ is the Doppler frequency set by vehicle speed, and $\beta_k$ is the large-scale gain of that hop. The detour also adds amplified surface noise, $v_k^H \Phi n_{\mathrm{RIS}}$, at the user. Here $\Phi$ is the diagonal matrix of surface phase shifts and amplification gains, and $n_{\mathrm{RIS}}$ is the thermal noise injected at the active elements.
Every quantity that sets the optimal common share is known in advance from sensor tracking: the predicted blockage start and end, which surface serves which user, the distances that fix $\beta_k$, the speed that fixes $f_D$, and the amplifier gain. The fast-varying small-scale fading is not known in advance, and the common message exists to absorb that uncertainty. The claim is falsifiable in two ways. If the episode-level plan does not come close to per-slot re-optimization, the claim fails. It also fails if robust beamforming without a common message matches the plan's performance during the detour window.
Research Program
The system is a multi-user 28 GHz highway downlink. A roadside base station serves several vehicles. A small number of active RIS units sit at pre-determined service regions, following the geometry of the pre-avoidance work. Trucks move as blockers under the 3GPP vehicle-to-everything (V2X) mobility and blockage model. The changing set of usable links (direct path, or detour through surface $j$) is represented as a sequence of quasi-static snapshots. This borrows the time-expanded graph idea from satellite-air-ground scheduling (Service Function Chain Dynamic Scheduling in Space-Air-Ground Integrated Networks), here at a single cell rather than across a network. Each snapshot specifies which links are open and how stale each user's channel estimate is.
The program has three parts. First, for one user on a detour, we derive how the optimal common-rate share and amplifier gain depend on $\sigma_k^2(\tau)$, the double-faded path gain, and the surface noise power. This includes imperfect successive interference cancellation, where a fraction of the common message leaks into private decoding. The ASTARS-RSMA analysis shows that this leakage drives diversity order to zero (Active STARS-Assisted Rate-Splitting Multiple-Access Networks), and that excessive amplification makes active surfaces worse than passive ones. We expect a coupling between these two effects: raising the amplifier gain raises the noise under which the common message must be decoded. If so, the best gain under heavy channel aging is lower than the best gain under perfect channel knowledge. Nobody has characterized this.
Second, we build the planner. At the sensing instant it takes predicted trajectories and outputs, for every snapshot in the episode, four things: the serving link per user, the common-rate share, the amplifier gain, and precoders designed on the statistics of the aging hop rather than its instantaneous value. The inner solver reuses the block coordinate descent of the active-RIS RSMA work for each snapshot, with the sum rate replaced by an ergodic rate under the error model above.
Third, we test whether the plan survives prediction errors in blocker position and in lane changes. Lane changes are the failure mode the predictor-antenna work reports.
Baselines
The naive reference is a blocked direct link with no surface, plus passive-RIS pre-assignment with single-user scheduling exactly as in Dynamic Blockage Pre-Avoidance using Reconfigurable Intelligent Surfaces. That paper reports that two passive units give a 30 dB transmit-power gain over the blocked link and recover 50% of the throughput loss caused by the truck. Three units come within 6% of adding a full-power base station. We reproduce its geometry and report the same "fraction of blockage-induced throughput loss recovered" metric so the numbers are directly comparable.
The strongest beamforming competitor is active-RIS beamforming without a common message (SDMA), designed robustly on the same channel statistics and given the same predicted serving schedule. It isolates whether the common message matters once robust design and prediction are both available.
The strongest optimization competitor is per-slot block coordinate descent active-RIS RSMA (Sum-Rate Maximization for Active RIS-Aided Downlink RSMA System). That paper reports up to 45% sum-rate gain over passive-RIS RSMA at equal energy and 8.34 to 15.68 bps/Hz as transmit power grows, converging in about 20 iterations. That was with perfect channel knowledge in a static setting, so its absolute numbers do not transfer to ours. We run it in two modes. In the oracle mode it gets fresh perfect channels every slot, which gives an upper bound. In the realistic mode it gets channels aged by the actual feedback delay.
The predictor-antenna scheme (High-Rate Uninterrupted Internet-of-Vehicle Communications in Highways) is a further baseline. It reports a 335% rate gain over outdated channel knowledge and a 20 dB outage-SNR gain, with proactive base-station switching. We implement it in our platoon scenario with a second base station available, and we ask whether surface detours with rate splitting add anything on top of it.
Two ablations isolate our ingredients. One freezes the common share at its pre-blockage value throughout the episode. The other plans the share but fixes the amplifier gain at its perfect-knowledge optimum.
All methods share the same channel realizations, sensor-prediction noise, total power budget (transmit power plus surface DC power), and solver iteration cap. Beating the field means meeting three conditions together:
- exceeding robust active-RIS SDMA on rate during the detour window by a margin outside seed-to-seed variation;
- reaching at least about 90% of oracle per-slot RSMA's episode-average rate at a small fraction of its computation;
- exceeding realistic-mode per-slot RSMA.
Why not just…?
Why not just use active-RIS beamforming without rate splitting, with predicted pre-assignment? This is the real threat. At high speeds the aging hop is nearly unknown, so robust beamforming might collapse to something close to broadcasting, and the common message would add little. The two-user analysis of RSMA (Rate-Splitting Unifying SDMA, OMA, NOMA, and Multicasting in MISO Broadcast Channel) predicts that beamforming without a common message plateaus as power grows under imperfect channel knowledge while RSMA keeps scaling. We test this directly by sweeping speed and power. If the gap is negligible across realistic highway speeds, we will report that the gain comes from prediction plus active amplification and that the common message is unnecessary.
Why not just re-optimize everything every slot? Per-slot re-optimization needs fresh estimates of a double-faded channel whose pilots arrive at low signal-to-noise ratio (Yanze Zhu et al.). It also needs about 20 solver iterations per slot. Our realistic-mode run measures what that schedule actually delivers once feedback delay is included. If it still wins, the planning idea is wrong.
Why not just avoid detours with predictor antennas and base-station switching? This works when a second base station is in range and vehicles hold their lane. The authors themselves note that it degrades under lane changes. Our prediction-error experiments include lane changes and single-base-station segments, which are the conditions where a surface detour is the only option.
Why not just add more passive elements? The pre-avoidance results cap passive recovery at about half of the lost throughput. Hardware results show active surfaces giving up to 130% capacity gains where passive surfaces give almost none (Active RIS vs. Passive RIS). We nonetheless include a passive arm with equal DC-power-normalized aperture, to avoid assuming the conclusion.
Proposed Experiments
The main experiment runs repeated truck-overtaking episodes in the 3GPP V2X 28 GHz model with Jakes-correlated small-scale fading. We sweep vehicle speed (which sets $f_D$), number of users, the number of active elements, and residual cancellation error. Metrics are the following:
- episode-average and worst-user rate;
- fraction of blockage loss recovered;
- outage probability within the detour window;
- energy efficiency including surface DC power;
- solver time per episode.
There are three decisive checks. The first plots the planned common-rate share over time against the oracle per-slot share. The claim requires that both rise in the detour window and track each other. The second compares the gap between rate splitting and robust beamforming without it, as a function of speed. The claim requires a gap that grows with speed. The third locates the optimal amplifier gain under aging and imperfect cancellation, against its perfect-knowledge value. The coupling hypothesis requires that the gain move downward. Robustness runs inject blocker-position error and lane changes to find where the episode plan breaks down, and compare the margin there against the predictor-antenna baseline.
References
- Dynamic Blockage Pre-Avoidance using Reconfigurable Intelligent Surfaces
- Sum-Rate Maximization for Active RIS-Aided Downlink RSMA System
- Active STARS-Assisted Rate-Splitting Multiple-Access Networks
- High-Rate Uninterrupted Internet of Vehicle Communications in Highways: Dynamic Blockage Avoidance and CSIT Acquisition
- A Primer on Rate-Splitting Multiple Access: Tutorial, Myths, and Frequently Asked Questions
- Service Function Chain Dynamic Scheduling in Space-Air-Ground Integrated Networks
- Rate-Splitting Unifying SDMA, OMA, NOMA, and Multicasting in MISO Broadcast Channel: A Simple Two-User Rate Analysis
- Channel Estimation by Transmitting Pilots From Reconfigurable Intelligent Surface
- Active RIS vs. Passive RIS: Which Will Prevail in 6G?