docs/07-rf-antenna.md
34 KB · 232 lines · markdown · View rendered · Download from the repository page
# 07 — RF & Antenna Plan **Source of truth:** `00-architecture-decisions.md` (ADR Baseline v0.3, 2026-09-27: **no built-in cellular modem** (confirmed final, D1); magnesium AZ91D chassis (D4 — RF-wise still a metal ground plane, §3); radios are Wi-Fi 7 on the MT7925 M.2 module and BLE 5.4 on the nRF5340 EC; antennas are 2× Wi-Fi flex in the front frame's upper band + 1× BLE chip antenna — ADR §2, §3.5, §5, §8, §9, §13 item 12, §14). This document is aligned to that revision; v0.1's cellular antenna plan (4× end-cap elements, SIM, carrier OTA) is superseded and summarised only in §2. **Feeds:** `08-mechanical.md` (RF window band, keep-outs), `09-pcb-architecture.md` (cable routing, shielding, clock spreading), `11-certification.md` (FCC/ISED/CE incl. Wi-Fi SAR), `12-risks-open-questions.md`.**Related:** `04-component-selection.md` (antenna part choices), `02-system-architecture.md` (radio roles, BLE on the EC), `06-thermal.md`. --- ## 1. Radio inventory and antenna count | Radio | Module / chip | Bands | Antenna ports | Connector | Notes ||---|---|---|---|---|---|| Wi-Fi 7 (802.11be) 2×2, 160 MHz; (BT on the module unused) | MediaTek MT7925 M.2 Key-E 2230 (e.g., AzureWave AW-XB591NF), `mt7925e` | 2,400–2,495 MHz, 5,150–5,895 MHz, 5,925–7,125 MHz | **2** (2×2 MIMO, tri-band shared) | 2 × MHF4 (I-PEX 20449) on the module | STA uplink + management AP (§9). Module BT is disabled (ADR §3.5) || BLE 5.4 | nRF5340 on the EC (always on) | 2,400–2,483.5 MHz | **1** | Chip antenna on the main board (Johanson 2450AT18x100 class) or PCB IFA fallback | Primary provisioning / control link, works with the SoC and Wi-Fi asleep || Cellular | **none** (ADR §14; see §2) | — | 0 | — | Cellular reach via phone hotspot (Wi-Fi STA) or a USB-tethered phone / USB cellular adapter on Port B || GNSS | none (ADR §13 item 10, closed D13: the iPhone app supplies location over BLE / NCM) | — | 0 | — | v0.1 used the modem's GNSS; no discrete receiver is planned | **Total: 3 antenna elements, 2 MHF4 cables** (v0.1: 7 elements, 6 cables). All behind PC (no metal) windows; no TPU over any window. --- ## 2. Why there are no cellular antennas The built-in 5G modem was removed in ADR v0.2 (`00-architecture-decisions.md` §14): the cellular uplink (2.5–10 MB/s in practice) was never something the hardware could fix, and the modem carried ~$112–146 of BOM, ~$100–150 k of one-time certification (PTCRB/GCF/carrier/cellular SAR), the post-DVT critical path, the hottest sustained heat source and the only rear-face part that forced 38 mm. Cellular reach is now provided by the user's phone — as a Wi-Fi hotspot to our STA, or USB-tethered on Port B — or by an off-the-shelf USB LTE/5G adapter on Port B (ADR §4.3, §5). That device carries its own radio grants, PTCRB/carrier approvals and SAR; PoppyUp is a USB host and network client of it, the same regulatory position as a laptop tethered to a phone (`11-certification.md` §5). Consequences for this document: the four wideband 600–6,000 MHz end-cap elements, their MHF4 cables and ferrite sleeves, the end-cap RF windows and their low-band volume protection, the flex-vs-LDS gate, cellular SAR and simultaneous-TX summation, cellular–Wi-Fi coexistence (n40/n41 vs 2.4 GHz, n77/n78 vs PCIe), the SIM tray and eSIM, and the TRP/TIS carrier OTA campaign are all deleted. What remains is a two-radio, three-element design whose only RF windows are in the front PC frame. The end caps are ordinary plastic (or, as an EVT option, magnesium as part of the chassis — `08-mechanical.md` §4). The removal is confirmed final (ADR §13 item 12, D1): no Key-B slot, SIM aperture or antenna window is carried. --- ## 3. The enclosure and the RF windows ADR §8: **magnesium (AZ91D die-cast) rear/bottom chassis-heatsink** (D4), hard-anodized aluminum sliding shutter over the top edge, PC+30 % GF front frame, TPU bumpers. Metal blocks everything; the antennas live in the plastic parts — **the front frame only**. RF-wise the change from aluminum to magnesium is neutral: magnesium's bulk conductivity is lower (~1/3–1/2 of aluminum's), but at 2.4–7 GHz the skin depth is a few µm and the casting remains a solid ground plane and shield. What matters is the coating (`08-mechanical.md` §4.1): the powder/e-coat top-coat is **non-conductive**, so every RF/EMI ground contact — the perimeter EMI gasket, the card-cage and shield-frame grounding points, the coax shield grounds near the antenna feeds — must land on **masked conversion-coated pads** (verify the conversion coat's contact resistance on coupons at EVT). ### 3.1 RF windows | Window | Material | Location (front-view frame, X from left, Y from top; `08-mechanical.md` §3.1) | Antennas | Size ||---|---|---|---|---|| Front-frame upper band, left zone | PC+GF 1.8 mm (Dk ≈ 3.2, tan δ ≈ 0.01), no TPU, no metallic finish | X 12–52, Y 30–38, between the slot row/LED bar and the battery door, left of the OLED | WIFI0 | 40 × 8 mm element zone; window clear area 44 × 10 mm || Front-frame upper band, right zone | same | X 130–170, Y 30–38, right of the OLED | WIFI1 | same; element centres 118 mm apart || Front frame, bottom-left | PC+GF 1.8 mm | X 11–18, Y 74–88, between the left end cap and the battery door aperture, above the corner bumper arm | BLE | 7 × 14 mm window over a 6 × 12 mm board keep-out (X 12–18, Y 75–87) || End caps (X 0–10, 180–190) | PC+GF (or magnesium, EVT option) | — | **none** | no RF requirement || Rear face, bottom face, top edge | magnesium chassis (coated) + hard-anodized aluminum shutter slats | — | none | continuous ground plane / shield; ground contacts only on masked lands | Design intent: the magnesium chassis (rear + bottom, ~190 × 96 mm) plus the reader-band top rail is a **single continuous ground plane** (the conversion-coated masked lands, not the powder coat, make the electrical connections); the Wi-Fi elements hang off its front perimeter along the upper band, exactly like laptop antennas along a lid edge. At 2.4 GHz (λ/4 = 31 mm) and above, a 40 mm flex element with the chassis as counterpoise is comfortably resonant — the low-band volume problem that drove v0.1's end-cap design does not exist any more. ### 3.2 PC window check at 1.8 mm (Wi-Fi and BLE) - Electrical thickness: at 7.125 GHz the wavelength inside Dk 3.2 is ≈ 23.5 mm, so 1.8 mm is ≈ 0.08 λ — a thin sheet, well below the first half-wave transparency/opacity cycle. Estimated transmission loss of a plane 1.8 mm PC+GF sheet: ≤ 0.2 dB at 2.4 GHz, ≤ 0.4 dB at 6–7 GHz (reflection-dominated; dielectric absorption ≈ 27.3·√Dk·tan δ·t/λ₀ ≈ 0.02 dB is negligible). Acceptable; no thinning of the window is needed.- **Verify Dk/tan δ at 2.4/5/6 GHz** — datasheet values for Lexan 3413 / Makrolon 9415 are quoted at 1 MHz–1 GHz; glass content moves Dk (30 % GF ≈ 3.2–3.4) and tan δ rises with frequency. Coupon test (stripline or split-post resonator) before frame tooling; tan δ ≤ 0.015 at 6 GHz is the acceptance limit.- Detuning: 1.8 mm of PC directly over a flex element shifts resonance down 2–5 %; matched in the enclosure at EVT (§4.4). Rain/sweat on the window shifts tens of MHz; the tri-band elements are broadband and ride through it — validate "wet" S11 at DVT.- Finish: the MT-11020 texture is RF-neutral. **No metallic-flake paint, chrome, EMI paint or conductive coating on the upper band or the BLE window**; the frame's conductive fabric-over-foam EMI strip stays inboard of the windows (`08-mechanical.md` §9).- No TPU over any window. Elsewhere the TPU grade still has to be non-conductive (no carbon black) so the bumpers on the front face's corners do not load the elements at X 12 and X 170: specify tan δ ≤ 0.03 at 2 GHz and confirm with the same coupon test. ### 3.3 Keep-outs | Rule | Value | Where it bites (front-view coordinates) ||---|---|---|| Wi-Fi element to any metal (chassis top rail, card cages, shutter slats, screws, OLED module frame, LED bar PCB, brass inserts) | ≥ 5 mm in the plane of the frame; ≥ 4 mm normal to the element | The card-connector housings begin ~4.5 mm behind the inner frame face (`08-mechanical.md` §3.3); the OLED module (X 64–126) is 12 mm from WIFI0 and 4 mm from WIFI1 — WIFI1's inner end may need to move to X 132 if EVT shows detuning; the shutter's closed slat 1 is ≥ 30 mm above the elements || Wi-Fi element to battery cell cans | ≥ 5 mm | Cells start at Y 44; elements end at Y 38 (6 mm) and sit in front of the pack in Z || Wi-Fi element to PCB components / other cables | ≥ 3 mm | LED-bar flex and OLED FPC route along the top of the band, not through it || Shutter detent magnets (N52) | ≥ 15 mm | Magnets at X 55 / 150, Z 33 on the top face; ≥ 30 mm from the elements || Fasteners within 8 mm of a Wi-Fi element | none ferrous — PC bosses with brass inserts or nylon screws | The 14 frame screws are on the perimeter and the door screws at Y ≥ 46 || BLE chip antenna | 6 × 12 mm copper-free keep-out on all layers, extending to the board edge, per the antenna vendor's reference layout; ≥ 3 mm to Port A's receptacle shell and bracket (X 20.5–29.5, Y 88–96) | The 360°-grounded shell is part of the ground plane the chip antenna needs, not a detuning body, but it is close: measure at EVT; fallback is to move the EC corner up to Y 60–72, further from the port row (`09-pcb-architecture.md` §4.2) || Antenna cables | ≥ 10 mm from buck inductors; ≥ 5 mm from the SoM LPDDR area | §4.3 | --- ## 4. Wi-Fi antennas ### 4.1 Layout (front view, top edge with slots at the top) ``` <-------------------------- 190 mm --------------------------> +-----------------------------------------------------------+ shutter/slot | CFast | CFe B | CFe A | SD | uSD | (top edge) | row (metal) +-----------------------------------------------------------+ end cap | LED bar ................................................... | end cap (plain PC, | [WIFI0 X 12–52] OLED [WIFI1 X 130–170] | (plain PC, rail groove | (X 66.5–123.5) Y 30–38 | rail groove + drain) |------------------- battery door X 18–172, Y 42–92 ---------| + drain) | [BLE] | | X 11–18 main board behind the pack; M.2 Wi-Fi X 140–162 | | Y 74–88 NVMe X 166–188 | +-----------------------------------------------------------+ | Port A Port B Port C RJ45 (bottom edge) | +-----------------------------------------------------------+``` Rationale for the upper band: the top edge is fully occupied by the slot row and covered by the aluminum shutter (open or closed); the bottom edge is full of connectors and their metal shells; the battery door cannot carry an antenna (cell cans 2.4 mm behind it); the rear face is the heat spreader and hosts the shutter pocket. The upper band is the only front-face strip that is plastic, at the ground-plane perimeter and never crossed by the shutter (§4.5). ### 4.2 Requirements | Item | Requirement | Note ||---|---|---|| Element | Tri-band flex-PCB monopole/IFA, 2.4 / 5 / 6 GHz (Taoglas FXP840 class; Molex 2118050001 / Antenova class as second source), ≈ 40 × 8–10 mm, adhesive-mounted on the inner face of the PC frame, MHF4 pigtail | `04-component-selection.md`. Both elements identical and both oriented along X for equal patterns || Bands | 2,400–2,495; 5,150–5,895; 5,925–7,125 MHz | Full Wi-Fi 7 tri-band incl. UNII-5–8 || Match | VSWR ≤ 2.5:1 (2.4 GHz), ≤ 2:1 (5/6 GHz) in the enclosure, both shutter states | 3-component π footprint (0402) at each MHF4 receptacle, tuned in the chamber at EVT || Peak gain | **≤ the MT7925 module's FCC modular-grant antenna-gain limits per band** (typically ≤ 5 dBi at 2.4 GHz, ≤ 6 dBi at 5/6 GHz — pull the AW-XB591NF grant and confirm) | Staying inside the grant keeps us a "host" of a modular transmitter; exceeding it means a Class II permissive change or our own 15C/15E radio test. A 40 mm flex on a 190 mm ground plane peaks at 3–5 dBi — inside the limit, but the measured pattern must be filed || Total radiated efficiency (in enclosure, free space) | ≥ −3 dB (50 %) at 2.4 GHz; ≥ −3 dB at 5 GHz; ≥ −4 dB (40 %) at 6 GHz, per element | Realistic for a flex element behind 1.8 mm PC on a chassis perimeter; drives the DVT throughput-vs-orientation test || Separation | **118 mm centre-to-centre** (X 32 and X 150) — 0.94 λ at 2.4 GHz, ≥ 2 λ at 5/6 GHz | `08-mechanical.md` §14 CTQ; minimum 60 mm || Isolation WIFI0 ↔ WIFI1 | ≥ 20 dB at 2.4 GHz, ≥ 25 dB at 5/6 GHz | Distance alone gives this; the OLED module between them helps || Envelope correlation coefficient | ≤ 0.1 all bands | For 2×2 MIMO gain and for the 2-stream uplink that carries the transfer || Shutter-state sensitivity | ≤ 1.0 dB efficiency shift, ≤ 2 % resonance shift open vs closed | §4.5 || Polarisation / pattern | Broad, front-hemisphere-weighted; no null toward the operator's phone (front) or the venue AP (any direction) | The metal (magnesium) rear face shadows the rear hemisphere by 6–10 dB — accepted; the device is used screen-up on a desk or cart (ADR §1) | ### 4.3 Cabling and routing - **2 × MHF4 (I-PEX 20449) 1.37 mm coax**, module → element. Loss ≈ 1.0 dB/m at 2 GHz, ≈ 2.2 dB/m at 5–6 GHz. - WIFI1: module (X 140–162, Y 14–44) → element at X 130–170, Y 30–38: ≈ 60 mm → ≤ 0.15 dB at 6 GHz. - WIFI0: module → element at X 12–52: ≈ 140–150 mm along the top rail → ≤ 0.35 dB at 6 GHz. Acceptable; if EVT wants the two chains balanced, the alternative is a longer WIFI1 loop, not a module move. - Budget used in §4.2 efficiency: 0.5 dB total including two connector interfaces (0.1 dB each).- MHF4 is rated ~30 mating cycles — fine for a factory-installed cable; the service note in `08-mechanical.md` §11 limits re-mating.- Routing: along the chassis top rail behind the reader band, then down to the frame's upper band at the reader-band/battery-band boundary; clipped every 30 mm (adhesive clips or cast-in channels) so drop shock cannot pull a connector; ≥ 10 mm from the buck inductors (bottom-left of the main board — the 5 V/6 A and 3.3 V bucks switch at 500 kHz–2.2 MHz with harmonics into the hundreds of MHz that couple onto a coax shield), ≥ 5 mm from the SoM's LPDDR area. Nothing is routed into the end caps.- Ground the coax shields to chassis ground at the element end via the small feed PCB (or the frame's EMI strip contact) within 10 mm of the element; the chassis-to-PCB ground connection (screws + gasket) must be RF-tight within 20 mm of each antenna feed (`09-pcb-architecture.md`). ### 4.4 Antenna technology | Technology | Verdict ||---|---|| **Flex-PCB tri-band element** (off the shelf, adhesive-mounted, MHF4) | **Plan of record EVT → MP.** ~$2 each; no tooling; matching network on our board is the only customisation; fastest path to real in-enclosure numbers || LDS (laser direct structuring) on the PC+GF frame | Optional PVT cost-down only (~$1.20/element + ~$15 k NRE): with two elements the payback is slow and it forces an LDS-grade resin on the whole frame. Decide at DVT; **not** a baseline requirement on the frame material (`08-mechanical.md` §4) || Chip antenna / Ignion-class booster for Wi-Fi | Not for the 2×2 uplink chains (efficiency); acceptable for the BLE element (§5) || Stamped sheet-metal elements | Only above ~50 k units/yr | ### 4.5 Metal shutter interaction - The shutter (two 2 mm aluminum slats) slides over the top edge into a pocket on the rear face; closed, slat 1 lies on the top face (Z 0–20) and slat 2 wraps the top-rear R10 corner; open, both are on the corner and the rear face. Because the Wi-Fi elements are on the **front** frame at Y 30–38, the shutter's travel never passes within ~30 mm of them, and in both states it is a fixed-length conductor bonded to the chassis ground — the detuning is small and static. (This is why v0.1 moved the Wi-Fi window from the rear strip to the front band; the reasoning stands.)- The 170 mm shutter (ADR §7) leaves 10 mm clear of each end cap in both positions; with no end-cap antennas in v0.2 this clearance is purely mechanical (rail grooves, `08-mechanical.md` §6).- Both states must be characterised in the chamber at EVT and DVT. The Hall sensor gives the EC the shutter state, but **no active aperture tuning** is implemented; if EVT shows a state-dependent shift > 1.0 dB on any band the fallback is a fixed compromise match. --- ## 5. BLE chip antenna - **Element:** 2.4 GHz ceramic chip antenna (Johanson 2450AT18x100 class, ~3.2 × 1.6 mm), on the nRF5340 EC section at the **bottom-left corner of the main board** (`09-pcb-architecture.md` §1.1), with the vendor's 6 × 12 mm copper-free keep-out running to the board edge and a 3-element π match. Fallback: a PCB IFA in the same keep-out (zero part cost, ~1 dB worse). Ignion-class booster is an option if the keep-out has to shrink.- **Window:** 7 × 14 mm clear PC+GF in the front frame at X 11–18, Y 74–88 — between the left end cap and the battery-door aperture, above the bottom-left corner bumper arm (TPU starts at Y 89); no TPU, no metallic finish (§3.2). Board keep-out X 12–18, Y 75–87.- **Targets:** total efficiency ≥ −4 dB in the enclosure; TX ≤ +3 dBm (nRF5340 maximum); RX sensitivity −95 dBm (1 Mbps) / −103 dBm (Coded S8). Range target: a reliable connection at 10 m in free space, and through a closed camera bag in transport (so the ADR §3.5 "Too hot — paused" notification reaches the app), with Wi-Fi STA active (§6, §10).- **Neighbours:** the cell cans start at X 23 (≥ 5 mm away, and behind the main board in Z); Port A's receptacle shell and bracket at X 20.5–29.5, Y 88–96 (≈ 3 mm from the keep-out corner) — the shell is 360°-grounded and acts as ground plane rather than detuning body, but this is the one placement risk in the BLE design and is measured at EVT with the fallback position (Y 60–72) reserved on the board outline. The three bucks bottom-left are ≥ 15 mm away and behind the board's ground pour.- **Role:** the BLE link is the **primary control path** (provisioning, wake, status, job control) and works with the SoC and Wi-Fi asleep; the Wi-Fi management AP is a convenience for bulk data (log pull, live view) once the app is already connected over BLE (ADR §3.5, §4.4). --- ## 6. Wi-Fi ↔ BLE coexistence in the 2.4 GHz band Both radios share 2.4 GHz, on different chips from different vendors and on different buses (MT7925 on PCIe to the SoC; nRF5340 on the EC). There is **no hardware PTA/coex line** between them; coexistence is spatial, spectral and temporal, coordinated by firmware over the SoC↔EC link. | Mechanism | Detail | Expected effect ||---|---|---|| Spatial separation | BLE element at X ≈ 15, Y ≈ 81; WIFI0 centre X 32, Y 34 (≈ 50 mm), WIFI1 ≈ 143 mm. Elements on different edges of the ground plane with the pack and door between them | Coupling ≈ −35 to −45 dB: a Wi-Fi TX at +20 dBm arrives at the BLE port at −15 to −25 dBm — a strong in-band blocker, so spatial separation alone is not enough at 2.4 GHz || Band choice | Wi-Fi STA prefers 5/6 GHz for the uplink; the management AP runs on 5 GHz only (§9); 2.4 GHz STA is used only when the venue or phone hotspot offers nothing else | In the common case there is no 2.4 GHz Wi-Fi transmitter in the box || BLE adaptive frequency hopping with a Wi-Fi-aware channel map | The SoC reports its 2.4 GHz channel and width (20/40 MHz) to the EC; the EC masks the BLE data channels under that footprint (≤ 9 of 37 for 20 MHz, ≤ 16 for 40 MHz — spec minimum 20 channels is kept in both cases) and re-enables them when Wi-Fi leaves the band. BLE's own AFH handles the rest | BLE never hops into the Wi-Fi channel; Wi-Fi TX bursts stop causing BLE packet loss, and BLE TX stops appearing as an in-channel blocker to Wi-Fi RX || Temporal | BLE connection events are ≤ 1 ms every 30–100 ms (< 3 % duty); the supervision timeout is ≥ 2 s; during 2.4 GHz Wi-Fi TX the EC shortens the interval to 30 ms and raises TX to +3 dBm so occasional collisions are lost events, not a dropped link. BLE is the controller's link, so the app's control channel degrades gracefully; Wi-Fi bulk data is the aggressor and the convenience path | Control link stays up under any Wi-Fi load || Filtering | MT7925 2.4 GHz band-edge filtering; BLE TX at +3 dBm couples ≈ −40 dBm into the Wi-Fi port — an in-band blocker only during BLE TX slots (µs), negligible | — || What we do not do | No wired PTA, no shared LNA switching, no BLE-over-Wi-Fi-module (BLE must work with the module powered down) | ADR §3.5 | Pass criterion (§10): BLE connection interval ≤ 100 ms with < 5 % lost events while Wi-Fi 2.4 GHz TX runs at full duty; Wi-Fi 2.4 GHz throughput loss < 5 % with BLE at a 7.5 ms interval. --- ## 7. EMI and self-desense | Source | Spectrum of concern | Victim | Mitigation ||---|---|---|---|| PCIe Gen3 (8 GT/s) on the SoC, switch, CFe B/A slots, GL9755/GL9750, RTL8126 | 100 MHz REFCLK harmonics (24th–25th at 2.4–2.5 GHz, 52nd–58th at 5.2–5.8 GHz), data-pattern spurs, radiated from the card edges | Wi-Fi 2.4/5 GHz RX, BLE | **RK3588 SSC on REFCLK** enabled in common-clock mode and forwarded by the ASM2812 (`09-pcb-architecture.md` §5): ~5–8 dB on the 100 MHz harmonics that fall in the 2.4 and 5 GHz bands — this, not cellular RX protection, is the v0.2 justification for SSC. Slot cages grounded to the chassis rail at multiple points; shutter closed during upload || PCIe Gen4 (16 GT/s) mux path and ASM2464PD (Port C, Pro) | 8 GHz fundamental; sub-rate clock spurs | 6 GHz Wi-Fi (UNII-5–8) | ASM2464PD SSC on; short traces on low-loss laminate; Port C at the far right of the port row, 25 mm from WIFI1's feed || USB4 20 GBd per lane (Port C) | Broadband to > 10 GHz; mostly above our bands | — | Shielded receptacle, 360° shell ground; primary 15B risk, not a desense risk || USB 3.2 Gen1 5 Gbps (Port A gadget) | 2.5 GHz half-rate content with SSC — classic **Wi-Fi 2.4 GHz / BLE** desense source | Wi-Fi ch 11–14, BLE upper channels | Shielded receptacle with 360° shell ground; USB SSC (mandatory in the spec) on; no common-mode chokes on SS pairs (SI); the BLE element is 3 mm from this shell (§5) — measured at EVT || USB 3.2 Gen2 10 Gbps (Port B host: SSD, tethered phone, USB cellular adapter) | 5 GHz half-rate content — sits at the low edge of UNII-1 (5.15–5.25 GHz) | Wi-Fi 5 GHz RX | Same shell grounding; SSC on; STA channel selection avoids ch 36–40 while Port B is active if EVT shows > 2 dB desense || RTL8126 5GbE PHY / RJ45 (line signalling to ~400 MHz; 125/156.25 MHz reference clocks) | 19th–20th harmonics at 2.375–2.5 GHz | Wi-Fi 2.4 GHz, BLE | Shielded MagJack with CMC magnetics grounded to the chassis, Bob Smith termination; the PHY sits top-right under the rear skin, ≥ 40 mm from any element; the Ethernet cable is the antenna for PHY noise — measured with the cable attached || LPDDR4x (2,133 MHz clock, 4,266 MT/s) | 2.13 GHz, 4.27 GHz, 6.4 GHz (3rd harmonic) | Wi-Fi 2.4 GHz band edge, 6 GHz band | SoM vendor lid shield; SoM ≥ 60 mm from either Wi-Fi element || Buck converters 500 kHz–2.2 MHz (harmonics to 200–300 MHz) | Conducted onto coax shields/pack cables | Nothing in-band; raises the noise floor at the module if picked up by the shields | Shielded inductors, tight input loops, coax routed per §4.3, spread-spectrum modulation on the bucks where available || Wi-Fi TX (2 × up to 20–23 dBm EIRP) into the rest of the system | Fundamental couples into everything | Card slots (a 20 dBm TX 30 mm from the CFast connector is ≈ −15 dBm at the card edge), USB gadget link, OLED | Shielded slots and cages; ESD/EMI ground fingers at the top rail; proven by the "radio-on" functional stress test (§10) | Shielding plan: a stamped sheet-metal shield frame over the SoM/switch/PHY area on the main board (also the thermal pad carrier) grounded to the chassis by at least eight screws; the reader daughterboard sits inside the magnesium top rail (grounded on masked lands), so the slots are effectively in a metal channel open only at the card openings. The battery pack cable is twisted and filtered at the charger with a common-mode choke. --- ## 8. Body, bag and RF exposure (SAR) ### 8.1 Body/bag loading- The device is operated on a desk, table or cart (ADR §1: a bag is for transport only); hand-held during card swaps and glances. A hand over the front upper band detunes 2.4 GHz by 20–40 MHz and adds 2–4 dB loss; the tri-band elements ride through it. Objects placed against the front face on a crowded cart (lenses, batteries, a laptop) can block one element — the 118 mm separation and 2×2 MIMO cover this; the link-rate falls, the link does not.- Cart/tripod mounting: the ¼-20 insert (`08-mechanical.md` §13) is on the bottom edge; metal plates there are ≥ 50 mm from any element. ### 8.2 SAR / RF exposure- **Classification: portable** (FCC KDB 447498 — used within 20 cm of the body: a user picks the device up while it is uploading). **The numeric SAR exclusion is not met by Wi-Fi:** the KDB 447498 test-exclusion criterion is (P[mW] / d[mm]) · √f[GHz] ≤ 3.0 at separations ≤ 50 mm; with the MT7925 at ~18–20 dBm (63–100 mW) per chain at 0–10 mm it evaluates to ≈ 10–15 at 2.4 GHz and higher at 5/6 GHz. So **SAR is tested, on the Wi-Fi bands only** (2.4 / 5 / 6 GHz), at 0 mm on the faces adjacent to the elements (front face, top edge) and at 10 mm as applicable, per KDB 447498 / 248227 / 616217 (`11-certification.md`). Only the BLE radio (≤ +3 dBm = 2 mW → ≈ 0.6) qualifies for the exclusion, so Wi-Fi + BLE simultaneous transmission (KDB 690783) is a paper exercise.- Two Wi-Fi chains transmit simultaneously (2×2). Their peak-SAR locations are 118 mm apart, so the 1 g volumes do not overlap and no summation applies; both elements are still each tested.- Expected result: 100 mW into a −3 dB-efficient element behind 2.5 mm of PC typically yields 0.3–0.8 W/kg (1 g) for laptop-lid-class Wi-Fi antennas — inside the FCC 1.6 W/kg (1 g) limit with margin; CE 2.0 W/kg (10 g) and the 4.0 W/kg extremity limit follow. **No proximity sensor / TX-backoff footprint is reserved** (v0.1 reserved one for the cellular end cap); if the DVT pre-scan disagrees, the MT7925 per-band power tables are the mitigation.- Cost: one module, three bands, no cellular bands and no cellular summation — roughly $12–20 k versus v0.1's cellular + Wi-Fi campaign (`11-certification.md`). RF-exposure statement and portable label text live in `11-certification.md`. --- ## 9. Wi-Fi operating rules (AP + STA concurrency, 6 GHz) - MT7925 supports STA + AP on the same channel (SCC) or with channel switching (MCC) with a throughput penalty; `mt7925` upstream supports 2 interfaces (verify current kernel).- **Default policy:** the management AP runs on **5 GHz** (UNII-1/3, 20/40 MHz, low power ≤ 14 dBm EIRP is enough for a phone at arm's length) on the *same channel as the STA uplink when the STA is on 5 GHz* (SCC — no loss); when the STA is on 6 GHz or 2.4 GHz, the AP still runs 5 GHz in MCC mode (the app link needs kilobits, so the ~10–20 % MCC airtime cost is acceptable and configurable). The AP is on only while the app is connected or for 5 min after a BLE "connect over Wi-Fi" request — **BLE is the primary control link; the AP is a convenience** (§5, §6).- **Phone-hotspot uplink** (ADR §4.3, §5): the STA joins the phone's hotspot like any other network; the link is tagged metered by default. iPhone hotspots are 5 GHz-capable (2.4 GHz if "Maximise Compatibility" is set) — the 2.4 GHz case is where §6 coexistence matters most.- **6 GHz AP is not offered**: in the US, 6 GHz APs are either *standard power* (requires AFC coordination — not for a portable device) or *low-power indoor* (LPI, which forbids battery-powered portable devices from operating as an AP; verify current FCC Part 15.407 wording); very-low-power (VLP) client/peer operation is allowed at ≤ 14 dBm EIRP and may apply to a future Wi-Fi Direct mode. The STA side uses 6 GHz freely where the venue's AP offers it.- Regulatory domain is set from the associated AP's country information element (802.11d), the app's location, or a user setting, falling back to the "world" domain (2.4 GHz + 5 GHz channels common to all). (v0.1 used the modem's network country; there is no modem.)- Wi-Fi Direct (P2P-GO) vs infrastructure AP: iOS does not support Wi-Fi Direct; the "device AP + app joins it" pattern (as used by cameras) is what we implement. iOS can be pointed at the AP programmatically via `NEHotspotConfiguration`. --- ## 10. Validation plan | Stage | Test | Setup | Pass criteria ||---|---|---|---|| PoC (month 1–2) | Ground-plane proof: two flex elements taped behind a 1.8 mm PC sheet on a 190 × 96 mm metal plate (aluminum is an adequate RF stand-in for the magnesium chassis) with a 3-D-printed front frame; passive S11 / efficiency | VNA + chamber rental (Satimo/MVG or CTIA-style) | Efficiency ≥ −4 dB at 2.4 GHz, ≥ −4 / −5 dB at 5 / 6 GHz before mechanical CAD is frozen || EVT | **Passive in-enclosure efficiency**: S11 (VSWR per §4.2), total efficiency, 3-D patterns and peak gain per band (for the grant check), isolation, ECC; shutter open and closed; dry and wet (sprayed); BLE element too | Anechoic chamber (rental, ~$3–5 k/day) with the machined AZ31B magnesium / printed-PC EVT enclosure (coated, with masked ground lands as in production) | §4.2 / §5 targets; both shutter states within 1.0 dB; peak gain ≤ module grant || EVT | **Conducted power at the module connectors**: MT7925 per-band/per-chain power tables and spurious at the MHF4 receptacles via RF-test-port variants (U.FL taps); nRF5340 conducted power | Spectrum analyser / callbox, shield box | Power tables match the module grant; cable loss ≤ 0.5 dB per §4.3 || EVT | **Radiated pre-scan**: FCC 15B/15C/15E and EN 301 489 spurious and band-edge with all data paths active | 3 m semi-anechoic pre-compliance | ≥ 6 dB margin before DVT tooling || EVT | **Desense**: Wi-Fi and BLE RX sensitivity (conducted, then radiated) with (a) idle, (b) CFe B read at 1.6 GB/s, (c) Port A gadget at 5 Gbps, (d) Port B SSD at 10 Gbps, (e) 5GbE active with cable attached, (f) USB4 on Port C (Pro), (g) all bucks at full load | Wi-Fi tester (Keysight / LitePoint / R&S) + shield box, then chamber | Desense ≤ 2 dB in any band under any load; if more, fix at EVT-2 before DVT tooling || EVT | **Coexistence read-under-TX**: BLE connection statistics (lost events, RSSI) while Wi-Fi 2.4 GHz TX runs at full duty on each channel with and without the EC channel map; Wi-Fi 2.4 GHz throughput with BLE at a 7.5 ms interval; card read in progress throughout | Device + phone + Wi-Fi 7 AP on 2.4 GHz at 3 m | §6 pass criterion; no data errors, no PCIe retrains, no USB re-enumeration during 1 h of all-radios-on stress || EVT → DVT gate | Antenna element and position frozen; LDS cost-down decision deferred to DVT | | || DVT | Wi-Fi throughput vs orientation, MIMO gain, AP+STA concurrency, phone-hotspot uplink | OTA with a Wi-Fi 7 AP and a phone hotspot at 3 m and 15 m; rotate the device | ≥ 80 % of conducted throughput at 3 m in all orientations; app link stable during STA uploads || DVT | **SAR pre-scan** (Wi-Fi 2.4 / 5 / 6 GHz, 0 mm on the front face and top edge, 10 mm; both chains) | SAR lab (~$8–12 k pre-scan) | < 1.6 W/kg (1 g) FCC with 15 % margin; no power backoff needed || DVT | BLE range and coexistence | 10 m free space, and through a closed camera bag (transport case) with Wi-Fi STA active | Connection interval ≤ 100 ms with < 5 % lost events || DVT | Tethered-uplink RF check | iPhone / Android tethered on Port B while Wi-Fi STA and BLE are active | No Wi-Fi/BLE desense from the phone's cellular TX at 0 mm on Port B (> 2 dB fails); informs the "keep the phone off the top face" user guidance || PVT | Production RF test | Per-unit conducted TX power / RX sensitivity via module test mode on the RF-test-port fixture, plus a quick OTA "golden unit" comparison in a shield box (couplers over the two Wi-Fi windows and the BLE window) | ±1.5 dB of golden | --- ## 11. Open items and notes against the ADR 1. **Module grant gain limits:** pull the AW-XB591NF (or chosen MT7925 module) FCC/ISED grant and confirm our measured peak gain per band is inside it; otherwise budget a Class II permissive change (`11-certification.md`).2. **PC+GF Dk / tan δ at 2.4–7 GHz:** coupon test before frame tooling (§3.2); acceptance tan δ ≤ 0.015 at 6 GHz.3. **BLE element vs Port A shell (3 mm):** EVT measurement with the fallback EC corner position (Y 60–72) reserved on the main-board outline (§5, `09-pcb-architecture.md`).4. **WIFI0 cable ≈ 150 mm:** accepted at ≤ 0.35 dB; confirm at EVT that the two chains are within 0.5 dB of each other, otherwise lengthen WIFI1's loop.5. **SAR test configuration:** agree 0 mm faces (front, top edge) and the 10 mm case with the lab early; it is Wi-Fi-only and sizes the DVT pre-scan (§8.2).6. **Port B 10 Gbps half-rate content near UNII-1:** include in the EVT desense matrix (§7); firmware channel-avoidance is the fallback.7. **`08-mechanical.md` §4:** the frame no longer needs an LDS-capable grade in the baseline; LDS is an optional PVT cost-down for the two Wi-Fi elements only.8. **End caps:** no RF requirement — the magnesium end-cap variant (ADR §8) has no RF consequence; the 10 mm shutter clearance at each end is a rail-groove CTQ, not an antenna CTQ (`08-mechanical.md` §14).9. **Magnesium chassis ground contacts (D4):** the powder/e-coat is non-conductive; confirm at EVT that the masked conversion-coated lands give a low, stable contact resistance for the EMI gasket, shield-frame screws and coax shield grounds, and that the 500 h salt-fog coupons (`08-mechanical.md` §4.1) do not degrade it.