RIS for 6G - real gains or academic hype?

Reconfigurable intelligent surfaces promise big coverage gains, but channel estimation overhead looks killer. Anyone seen convincing real-world RIS measurements? Or is it still papers only?

Answering the “papers only?” ask with what measurement campaigns have actually shown (habits — retune to *your* band, aperture size, and CSI rate):

1. **What has been measured (not brochure)**
- Controlled **indoor / corridor / factory** demos at sub-6 and some mmWave bands have shown **several dB to low-teens dB** coverage / SNR lift in the shadowed zone when the RIS is sized, placed, and configured for that geometry. That is real.
- The gain is **highly geometry- and state-dependent**: wrong panel orientation, blocked Fresnel, or stale phase map → the lift collapses. Chamber “N×N RIS, perfect CSI” curves do **not** transfer to a hallway with people walking.
- Outdoor / street-canyon results exist but are fewer, often with a known UE location and a slow control loop — not yet a drop-in for a live cellular scheduler.

2. **Where the overhead actually bites**
- **Channel estimation / CSI:** a large passive RIS has no RF chains; you estimate the cascaded Tx–RIS–Rx channel. Pilot / sounding cost scales with the number of controllable elements (or groups) you want to resolve. At high mobility or multi-user, the CSI age kills the beamforming gain before the panel does.
- Practical demos that look good usually use **grouped elements, codebook / DFT beams, or a slow quasi-static map** — not per-element, per-TTI perfect CSI.
- Hardware: PIN / varactor / liquid-crystal / MEMS panels each bring **insertion loss, phase quantization, and bias-network RF leakage**. A 1-bit panel still works for coverage fill; it will not match a continuous-phase paper plot.

3. **How we read a “convincing” RIS paper / demo**
- Same UE location **with RIS off / random / optimized** on the same day, same TX power, same antenna.
- Report **net** SNR / throughput after sounding overhead, not only peak RSSI with frozen CSI.
- State panel IL, bit depth, update rate, and whether the controller had **side information** (UE position) that a real network would not.

So: **not pure hype** for coverage fill / NLOS rescue in a controlled geometry; **still research-grade** for dynamic multi-user cellular where CSI and mobility dominate. Open: are you targeting a **quasi-static indoor fill** (factory / lobby) or a **mobile outdoor** cell — and roughly how many controllable elements / update rate can your control link afford?

— rfforum 编辑部 / Editorial Team