Make your paper interactiveCreate

Text from the interactive explanation

Reconfigurable Intelligent Surface Relying on Low-Complexity Joint Sector Non-Diagonal Structure

Yinuo Dong, Qingchao Li, Soon Xin Ng, Mohammed El-Hajjar

Original paper

Dong, Li, Ng & El-Hajjar · IEEE Open Journal of Vehicular Technology · 2024Dong et al. · IEEE OJVT · 2024

A six-sided smart surface that could catch a phone signal on one side and hand it to everyone around it.

Drag the walker anywhere, or use the slider. Then switch how the pillar serves you.

Serving faces 4 · 5 · 6

angle factor (paper's pathloss model, α = 2.5); the tick marks 1.00 = one face looking straight at you

0° = far side, behind the pillar; 180° = tower side. Pause freezes waves, pulses and walking.

You're near a corner: the side face is pulling its weight.

Illustrative wave picture built from the paper's model, not a measurement. Teal ripples: the tower's signal reaching faces 1–3. Glowing traces on the pillar's top: switches passing that captured signal to the faces that look at you. Amber ripples: re-sent by those faces, timed to arrive in step at your phone. Tile colours show each tile's phase setting.

Chapter 1

Tiles that delay the echo

Start with one flat panel. A reconfigurable intelligent surface , or RIS, is a sheet tiled with many cheap, passive elements. None of them amplifies anything. Each one simply re-radiates the radio wave that hits it, with a small adjustable delay — a phase shift.

Your phone hears the sum of all those echoes. If they arrive in step, crest on crest, they add up; engineers call this coherent combining. If they arrive out of step, they cancel each other out.

Line the arrows up by hand — then let the surface do it.

Each button below is one tile. Press it to delay that tile's echo by another 45°. Its arrow shows how that echo arrives at the phone.

Received power: 1.0× one tile

What you're seeing: the tiles are all set to the same delay, so they act like a plain mirror: their echoes reach the phone out of step.

Auto-tune applies the paper's rule (its Eq. 29): each tile's phase is set to cancel the delay of its own two-hop path, tower → tile → phone, written θi = −(∠hi + ∠[Gw]i). Once every echo is aligned, the chain is as long as the number of tiles, N, and the power is N² times one tile: eight tiles give 64×. Double the tiles, four times the power. Drag the phone after auto-tuning and the surface keeps re-aiming.

Illustrative model: every tile's echo is drawn with equal strength, so only the timing matters. Real tiles also differ in how strongly they catch and throw — that difference becomes useful in Chapter 5.

Chapter 2

A mirror only works for people in front of it

A flat surface helps only the half of the world it faces. Walk behind it and you are in its shadow. The two-sided STAR-RIS (simultaneously transmitting and reflecting) also serves the back, but not the edges: seen edge-on, any face is almost useless.

How useless? In the paper's pathloss model the strength a face can deliver falls like cos2.5φ, where φ is the angle between your direction and the face's normal, the line pointing straight out of it. At 60° off the normal a face gives 18% of its head-on strength; at 90°, nothing.

Add surfaces one after another, then walk all the way round (drag on the plot or use the slider).

Angle factor here: 0.70 — good

What you're seeing:

Angle factor only; total strength also depends on how many tiles serve you — the paper's full simulations (below) compare equal tile counts.

Two things shrink as you move off to the side. The face looks narrower from where you stand (its projected area falls like cosφ), and each tile radiates mostly straight ahead rather than sideways. The paper's pathloss model (its Eqs. 9–11) multiplies a radiation pattern, F = 2(α+1)·cosαφ with α = 2.5, which is zero beyond 90°, by the face's projected area. The angle factor on this page is the cos2.5φ term for each face, the same per-face term the paper compares in its directional-gain analysis (Eq. 62).

Wrap six faces into a hexagonal pillar and every direction has a face within 30° of square-on: even at a corner the best face keeps 0.70 of its full strength. That is the paper's multi-sector RIS . The tower (the base station) stands on one side, squarely facing face 2, and the users can be anywhere around the pillar.

Look at the paper's drawing. The tower's rays (blue) land on the faces turned toward it, sectors 1–3. User k's rays (red) leave from the faces turned toward them. On the far side of the pillar those are different faces, and that raises two questions.

Should a user be served by just one face, or by several at once? Chapter 3 lets you measure the answer.

How does a signal caught on the tower's side reach faces on the far side? Chapter 4 opens the pillar and shows the switches.

Chapter 3

Faces that team up

An earlier multi-sector design served each user with one face — whichever looks most directly at them. The paper's first idea is to serve every user with half the pillar at once : with six faces, the three adjacent faces turned toward you (the central face plus its two neighbours), no matter where you stand.

Predict first. Where does teaming up help most, compared with the single best face?

Pick one, then drag the slider to find out.

Gain of three faces over the best single face

+35%

What you're seeing: each bar is a face's angle factor, cos2.5 of its angle to you. All the way round the pillar (curve) the gain swings between +35% in front of a face and +100% at every corner — the “30%−100% gains as the users' positions change” reported in the paper's abstract.

The paper puts the two extremes like this. Straight in front, “the bilateral side sectors can provide less than one-fifth of the performance enhancement relative to the central sector”. At the 30° edge, “the nearer side sector can provide performance support equivalent to that of the central sector”. A corner, the worst spot for a single face, is exactly where teamwork pays most.

Chapter 4

Passing the signal round the back

Only faces 1, 2 and 3 can see the tower. To serve someone on the far side, the signal they catch must be handed to other faces. The paper does this with switches inside the pillar , and proposes two ways to wire them (horizontal connections between faces).

Opposite-link (Structure 2): each capture face is wired only to the face opposite it — 1↔4, 2↔5, 3↔6. You are served by the fixed group 1·2·3 or 4·5·6, whichever contains the face nearest you.

Full-link (Structure 4): any capture face can feed any face, so you always get the best trio centred on you.

Choose a wiring, then drag the walker round the pillar (or use the slider).

What you're seeing: a schematic, one line per bundle of switches. Teal faces catch the tower's signal; closed switches glow and pulses carry the signal to the amber faces that serve the walker. Switch counts come from the paper's Table 3 formulas for its Figure 1 example (six faces, three tiles each, N = 18): opposite-link needs Nx·Ny = 18, full-link Nx²·Ny = 108.

Stand in front of face 4 and the difference shows: opposite-link must use 4·5·6 (angle factor 1.18), while full-link picks 3·4·5 (1.35). Opposite-link trades a little flexibility for a much simpler circuit. The paper also notes that switches use less energy than the tunable impedance components of other designs, and that RF MEMS switches offer near-zero power consumption at low cost.

Chapter 5

Swap partners

The second idea works inside each face (vertical connections). Tiles are not equal: because of fading, some catch the tower's signal strongly and some weakly, and the same goes for how strongly each tile's echo reaches you. In an ordinary “diagonal” RIS each tile re-sends what it caught. In the paper's non-diagonal design, switches let a group of Gs tiles swap partners: the signal caught by one tile can be re-sent by another.

The rule (from earlier work the paper builds on) is simple: sort the catch strengths and the throw strengths, then wire the strongest catcher to the strongest thrower, second to second, and so on. Phases are still aligned, so the amplitude at your phone is the sum of catch × throw products — and sorted pairing makes that sum as large as it can be.

One face, six tiles (Gs = 6). Tap a catcher, then a thrower, to wire them together (their old partners swap). Or let the rule do it.

+0% power vs. straight-through

What you're seeing: line thickness is each pair's catch × throw product. The bar adds them up (amplitude at your phone); power is its square. Marks show straight-through and the best possible wiring.

Does it pay off on average?

One lucky draw proves little, so roll thousands of random worlds. For random (Rayleigh-faded) channels the paper derives the average power of straight-through wiring, N + N(N−1)·π²/16 (its Eq. 53), and shows that sorting within big groups approaches N² (Eq. 58) — the bound of a fully connected surface. That is an asymptotic advantage of 16/π² ≈ 1.62, or +62%.

Pick a surface size and a group size, then run.

Press run to simulate.

Simplified in-page simulation of the paper's Section V analysis (single antennas, random Rayleigh channels, phases aligned). Average power in units of one tile's average power.

More freedom costs more switches

Bigger groups give the sorting more room, so more gain — but every extra partner needs switches. Here is the trade-off for one example: six faces of six tiles (N = 36), opposite-link wiring plus swapping (Structure 3).

Power gain: simplified in-page simulation (4,000 random worlds each). Switches: from the paper's Table 3 formulas, Nx·Ny + Gs·Ny with Nx = Ny = 6 (group size 1 means no swapping, i.e. plain Structure 2).

Chapter 6

What the simulations showed

The paper's results come from theory and Monte-Carlo computer simulation (105 random trials), not from measurements of a built pillar. The simulated set-up: a base station with M = 4 antennas, 51 m from the pillar, transmitting 50 mW; noise at −90 dBm; tiles half a wavelength across; users roaming on a circle 15–50 m out; perfect channel knowledge unless stated. Rates are in bps/Hz, bits per second per hertz — how much data each slice of radio spectrum carries.

The six designs, side by side

Click a row to see its wiring on the mini pillar (user near the corner of faces 4 and 5).

Pillar versus STAR-RIS versus a flat panel

Achievable rate at N = 54 tiles (bps/Hz, approximate, read from Fig. 12)

The simulations also report that Structure 5 beats Structures 4 and 1 for every Rician factor, element size and distance tested, with the gain shrinking as the line-of-sight share of the signal grows (Figs. 9–11), and that the proposed design stays ahead with imperfect channel estimates (Fig. 13).

Chapter 7

Back to the square

Now fill the square. One tower stands on one side; a pillar of tiles stands where people pass — the authors suggest “shopping malls, stations, and public squares”, “enabling a single-side deployment of the BS that can serve multiple users omni-directionally”, and see busy central locations helping vehicular communications too.

Three walkers, each served by its own trio of faces.

What you're seeing: an illustrative superposition — each walker gets its own three faces and its own focal spot. The paper's multi-user result is the simulation in Figure 17 above.

Everyone around the pillar is served, from a tower on one side. And the design becomes a dial between performance and circuit simplicity: Structure 3 (opposite-link plus swaps) when switches must stay cheap, Structure 5 (full-link plus swaps) when every bit per second counts.

Most results assume perfect channel knowledge; one simulation (Fig. 13) adds estimation errors.

Far-field pathloss model with the angle exponent α = 2.5, with all signals treated as reflected from the pillar's centre.

A single-cell downlink: one base station sending to users; the base station always talks to faces 1–3.

Results come from theory and Monte-Carlo simulation of the proposed design, not from hardware.

Source

Yinuo Dong, Qingchao Li, Soon Xin Ng and Mohammed El-Hajjar, “Reconfigurable Intelligent Surface Relying on Low-Complexity Joint Sector Non-Diagonal Structure”, IEEE Open Journal of Vehicular Technology, vol. 5, pp. 1106–1123, 2024. DOI: 10.1109/OJVT.2024.3447109

Figures reproduced from the paper under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/), cropped. Interactive visuals are original illustrations built from the paper's models; the in-page Monte-Carlo is a simplified simulation of its Section V analysis.

Typefaces: Space Grotesk, Newsreader and IBM Plex Mono, used under the SIL Open Font License 1.1.

Topics in this iPaper

Engineering
we design a multi-sector RIS joint service model based on the BD RIS architecture and we compare the corresponding system rates, circuit complexity, and gains provided by different architectures

Abstract, page 1106

Computer science
a joint beamforming design for maximizing the achievable sum-rate is presented

Section IV, Beamforming Design, page 1114

reconfigurable intelligent surfaces
RIS is an artificial two-dimensional (2D) surface equipped with a large number of low-cost passive components

Section I, Introduction, page 1107

beamforming
the beamforming vector w can be designed using Maximum Ratio Transmission (MRT) beamforming method

Section IV-A, page 1114

wireless communications
we consider the RIS-aided downlink MISO and MIMO communications in a single-cell network

Section II, System Model, page 1109

multiple input multiple output
In a multi-user MIMO system, the BS has M antennas communicating with K single antenna users.

Section IV-B, page 1115

semidefinite relaxation
We can apply the SDR to solve this problem.

Section IV-B, page 1115

Reconfigurable Intelligent Surface Relying on Low-Complexity Joint Sector Non-Diagonal Structure — Interactive explanation | iPaper.ai