Digest · Measuring Starlink · read after Chapter 3
導讀 · 量測 Starlink · 第 3 章後閱讀
Democratizing Direct-to-Cell Low Earth Orbit Satellite Networks讓直連手機的低地球軌道衛星網路普及化
This digest explains how orbital motion disrupts cellular session state and how MOSAIC uses locally verified tokens to reduce signaling. It also identifies the unmodeled ISL assumptions behind its reported gains.讀完後，你能說明為什麼一個在地面上逐跳建立的蜂巢式會談，在基地台是以軌道速度移動的衛星時會失效，MOSAIC 如何以一次性權杖取代這種會談，讓自給自足的衛星能自行驗證，以及它那些大幅加速中，哪些依賴論文模型範圍之外的星間鏈路假設。
Main result
主要結果
MOSAIC moves cellular admission to the satellite through pre-issued, locally verified tokens. Its prototype and emulation examine signaling, coverage and setup under stated access assumptions.
MOSAIC 以預發且在本地驗證的權杖，將蜂巢接納決策移到衛星，其原型及模擬在明列接入假設下評估信令、覆蓋與建立流程。
| Item項目 | Values數值 | Conditions and source條件與來源 |
|---|---|---|
| Session state | Qualitative scope | Liu et al. examine direct-to-cell service, where an ordinary phone connects to an overhead satellite. They identify session state as an architectural constraint. Each session links state across the phone, radio, satellite operator, and mobile operator’s core. |
| Evidence and scope | Qualitative scope | so a satellite that serves separate operators' customers every few minutes must constantly rebuild sessions and exchange signalling with distant cores. |
| Orbital configuration | Qualitative scope | MOSAIC places the radio and user plane on each satellite. Mobile operators pre-issue policy-carrying, one-time tokens with blind signatures. The SIM enforces single use, and the phone sends tokens in-band for local access to an overhead satellite. |
| Dataset and workload | 50.9 · 943,003 · 50 · 3 · 116% · 5.19 · 850–7,640 | The evaluation combines 50.9 million TLE records, 943,003 Iridium RF measurements, 50 RIPE Atlas Starlink probes, and an Amarisoft 3GPP NTN prototype. Commercial phones participate in the prototype. Reported gains include up to 116% more serviceable area and up to 5.19× faster Starlink service setup. Inter-operator handover signaling falls by 850–7,640× against 3GPP NTN [59, §7, Figs. 15, 17, 21]. Tsinghua reports that the paper received an Outstanding Paper Award. |
| Latency comparison | 20 | These results come from data-driven emulation and a ground prototype. The paper cites a typical 20 Gbps ISL. ISL latency, topology, and setup remain outside the model. Measured direct-to-cell latency would provide an external baseline. |
| 證據與範圍 | 定性範圍 | Liu et al. 主張，直連手機的衛星服務（一般手機直接與上方的衛星通訊）受限的蜂巢網路保存狀態的方式，射頻物理則是另一組限制：每一個會談都在手機、射頻、衛星營運商與行動營運商核心網路之間逐跳拼接，所以每隔幾分鐘就服務相異營運商客戶的衛星，必須反覆重建會談並與遙遠的核心網路交換信令。 |
| 證據與範圍 | 定性範圍 | 他們的設計 MOSAIC 讓每顆衛星自給自足，在星上執行射頻與使用者平面，並讓行動營運商預先發出帶有政策的一次性權杖，以盲簽章簽署、由 SIM 卡強制只能使用一次，手機在頻內花用它，向當下位於上方的任何衛星購買在地存取。 |
| 資料集與工作負載 | 5,090 · 943,003 · 50 · 3 · 116% · 5.19 · 850–7,640 | 以 5,090 萬筆 TLE 紀錄、943,003 筆 Iridium 射頻量測、50 支 RIPE Atlas Starlink 探針，以及搭配商用手機的 Amarisoft 3GPP NTN 原型，論文報告可服務面積最多增加 116%、在 Starlink 上服務建立最多快 5.19 倍、跨營運商換手信令比 3GPP NTN 低 850–7,640 倍[59, §7, Figs. 15, 17, 21]，並依清華大學的公告獲得傑出論文獎。 |
| 容量與速率 | 20 | 這些結果來自資料驅動的假設情境模擬，，在軌衛星的測試則待完成，它們引用「典型」的 20 Gbps ISL，卻ISL 延遲、拓樸與建立時間列於模型範圍之外，而且實測直連手機延遲仍待蒐集，才能作為比較基準。 |
1. Where this paper sits1. 這篇論文的位置
The paper extends Earth-fixed geography from routing to cellular access. It changes session-state handling while using the existing NTN standards context.
本篇將地球固定地理從路由延伸到蜂巢接入，在既有 NTN 標準背景下改變會談狀態處理。
| Item項目 | Values數值 | Conditions and source條件與來源 |
|---|---|---|
| Evidence and scope | 3 · 2 · 5 · 0 · 4 | Most papers in the Measuring Starlink group study dish-based broadband. MOSAIC examines access for ordinary phones, which Chapter 3 calls direct-to-cell. Starlink uses 2 × 5 MHz of T-Mobile’s PCS spectrum for this service. Crowdsourced outdoor measurements show median SINR near 0 dB and roughly 4 Mbps per beam [58]. |
| Evidence and scope | Qualitative scope | Orbital elements, Iridium RF traces, and Starlink probes support MOSAIC’s problem statement. Its contribution is an access architecture. It connects to routing papers that use Earth-fixed geography in place of satellite identity. |
| Evidence and scope | Qualitative scope | SHORT applies Earth-fixed geography to routing tables, while TinyLEO applies it to anycast. MOSAIC applies it to access. Each satellite admits users through valid tokens within Earth-fixed geographic cells. |
| Cellular standards | 3 · 17 · 18 | MOSAIC builds on 3GPP NTN Releases 17 and 18, which extend cellular architecture to satellites and retain its session model. MOSAIC changes that session model. |
| 證據與範圍 | 3 · 2 · 5 · 0 · 4 | 「量測 Starlink」組的多數論文看的是透過終端天線提供的寬頻服務，而這篇看的是能觸及一般商用手機的較新服務，第 3 章稱之為直連手機，Starlink 目前在 T-Mobile 的 2 × 5 MHz PCS 頻譜上提供，群眾外包量測顯示其中位數 SINR 約 0 dB、戶外每波束約 4 Mbps[58]。 |
| 證據與範圍 | 定性範圍 | 它歸在這一組，是因為其問題陳述建立在量測上，來自軌道根數、Iridium 射頻軌跡與 Starlink 探針，但它的貢獻是一種架構，與那些以地表固定地理取代衛星身分的路由論文同屬一類。 |
| 證據與範圍 | 定性範圍 | SHORT 對路由表這樣做，TinyLEO 對任播這樣做，而 MOSAIC 對存取這樣做，使用地表固定的地理服務格，以及一顆只需取得使用者的有效權杖的衛星。 |
| 證據與範圍 | 3 · 17 · 18 | 它的標準背景是 3GPP 在 Release 17 與 18 中的衛星接入網路工作，那些工作把蜂巢架構延伸到衛星，但保留了其會談模型，而 MOSAIC 的核心主張正是這個會談模型才是該改變的東西。 |
2. The problem, measured2. 問題，用量測說話
Orbital motion repeatedly changes the operators and users that a satellite serves. Hop-by-hop session state then generates core signaling, while radio deadlines favor on-board processing.
軌道運動反覆改變衛星服務的營運商與使用者，逐跳會談狀態因而產生核心信令，射頻期限則有利於機上處理。
| Item項目 | Values數值 | Conditions and source條件與來源 |
|---|---|---|
| Evidence and scope | Qualitative scope | The problem is a mismatch between who owns what and how fast it moves. |
| Operator coordination | Qualitative scope | The satellite operator owns the spacecraft. The mobile operator manages subscribers and the core that authorizes, bills, and routes their traffic. Conventional sessions link state across these systems. A change of serving satellite changes that chain and requires close operator coordination [59, §3]. |
| Capacity and rate | 340–2,000 km · 15.75% · 20 | Satellites at 340–2,000 km cross regions within minutes. Satellite, operator, and user relationships therefore change frequently across many parties. The authors associate this coupling with service-area restrictions and signaling storms. In their model, signaling consumes up to 15.75% of a typical 20 Gbps ISL [59, §3.1–3.2, Fig. 7]. |
| Evidence and scope | 250 µs | A transparent satellite relays radio signals to a ground base station. Cellular scheduling deadlines near 250 µs require a shorter path than that ground round trip [59, §3]. |
| Dataset and workload | 50,922,755 · 943,003 · 7.8 · 382,000 · 50 · 3,903,065 | The empirical base contains 50,922,755 TLE entries and 943,003 Iridium RF measurements. It also includes 7.8 million pings and 382,000 traceroutes from 50 Starlink RIPE Atlas probes. A GEO direct-to-device service contributes 3,903,065 logs [59, Fig. 3]. |
| 證據與範圍 | 定性範圍 | 問題在於誰擁有什麼與它移動得多快之間的差異。 |
| 證據與範圍 | 定性範圍 | 衛星營運商擁有太空載具，行動營運商擁有用戶以及負責授權、計費與路由的核心網路，而在現今的設計中，手機的會談是一串保存在兩者之間每一跳的狀態，所以兩家營運商必須緊密耦合，每次更換服務衛星都會變成那串狀態的改變[59, §3]。 |
| 容量與速率 | 340–2,000 km · 20 · 15.75% | 由於 340–2,000 km 的 LEO 衛星在幾分鐘內就經過一個地區，衛星、行動營運商與使用者之間的關係以多對多的方式持續變動，在作者的分析中，這限制了某營運商能提供服務的地方，並產生信令風暴，當信令經由典型的 20 Gbps ISL 傳送時，最多可佔用星間鏈路頻寬的 15.75%[59, §3.1–3.2, Fig. 7]。 |
| 證據與範圍 | 250 µs | 顯而易見的出路，也就是只把射頻訊號轉送到地面基地台的透明衛星，則因另一個原因失敗，因為蜂巢射頻約 250 µs 的排程期限需要比往返地面更短的路徑才能達成[59, §3]。 |
| 資料集與工作負載 | 50,922,755 · 943,003 · 50 · 780 · 382,000 · 3,903,065 | 作者把這些論點建立在龐大的實證基礎上：50,922,755 筆 TLE、943,003 筆 Iridium 射頻量測、來自 50 支裝在 Starlink 終端上的 RIPE Atlas 探針的 780 萬次 ping 與 382,000 次 traceroute，以及 3,903,065 筆 GEO 直連裝置服務的紀錄[59, Fig. 3]。 |
3. The design3. 設計
Blind-signed tokens let satellites verify local access while retaining subscriber anonymity. SIM-enforced single use supports handover, with delayed reuse reporting on ground-only satellites.
盲簽章權杖讓衛星在本地驗證接入，同時保留用戶匿名性。SIM 強制單次使用以支援換手，僅用地面鏈路的衛星則延後回報重用。
| Item項目 | Values數值 | Conditions and source條件與來源 |
|---|---|---|
| Evidence and scope | Qualitative scope | MOSAIC moves admission decisions from the remote mobile core to the overhead satellite. Tokens carry the authorization needed for this local decision. |
| Evidence and scope | Qualitative scope | The mobile operator issues each subscriber a supply of one-time tokens in advance, each carrying the policy that governs its use, and signs them blind. Thus, that the satellite operator can later verify a token's authenticity while retaining subscriber anonymity [59, §4–6]. |
| Evidence and scope | 3,279 · 19,661 | The SIM stores the tokens and enforces that each is spent only once. This explains why the scheme can be deployed incrementally by changing SIMs rather than phones. A SIM holds between 3,279 and 19,661 tokens depending on the configuration [59]. |
| Orbital configuration | Qualitative scope | When a phone wants service, it presents a token in-band to whichever satellite is overhead. The satellite, which runs the radio and user plane itself and is therefore self-contained, verifies the token locally and grants pay-as-you-go access through local verification. |
| Evidence and scope | 0.2 · 0.88 · 22.8 · 87.7 | Spent tokens require later checks for reuse. A satellite with ground links reports them when it next reaches a ground station. Detection can therefore take 0.2–0.88 orbital periods, or 22.8–87.7 minutes [59, Fig. 20]. |
| Local handover | Qualitative scope | Earth-fixed cells retain their boundaries as satellites pass. Local verification supports handover between satellites, including satellites owned by different operators. |
| Evidence and scope | 1,175 · 1,401 · 16.4 · 41.9 · 34.8 ms | One CPU core generates about 1,175 tokens per second and verifies 1,401. Maximum spending latency is 16.4, 41.9, or 34.8 ms across the three studied constellations [59, §7.3]. |
| 證據與範圍 | 定性範圍 | MOSAIC 把「使用者如何可以被服務」的決定，從經由網路抵達的行動營運商核心網路，移到經由空中抵達的上方衛星，而權杖讓這個移動變得安全。 |
| 證據與範圍 | 定性範圍 | 行動營運商事先給每位用戶一批一次性權杖，每個都帶有規範其使用的政策，並以盲簽章簽署，使衛星營運商之後能驗證權杖的真實性，同時保留用戶匿名性[59, §4–6]。 |
| 證據與範圍 | 3,279 · 19,661 | SIM 卡儲存權杖並強制每個只能花用一次，這就是為什麼此機制能以更換 SIM 卡、保留原手機的方式逐步部署，而一張 SIM 卡依設定可容納 3,279 到 19,661 個權杖[59]。 |
| 證據與範圍 | 定性範圍 | 當手機需要服務時，它在頻內向當下位於上方的任何衛星出示權杖，而衛星自己執行射頻與使用者平面，因此是自給自足的，它在本地驗證權杖並授予隨用隨付的存取，以本地驗證完成授權。 |
| 證據與範圍 | 0.2 · 0.88 · 22.8 · 87.7 | 已花用的權杖最終必須檢查如何被重複使用，而僅有地面鏈路的衛星只能在下次接觸地面時回報，所以濫用可能在 0.2 到 0.88 個軌道週期內等待偵測，也就是 22.8 到 87.7 分鐘[59, Fig. 20]。 |
| 證據與範圍 | 定性範圍 | 服務格定義在地表，衛星則負責服務該格子波束上，所以使用者的服務格在衛星經過時保持固定，而把使用者從一顆衛星交給下一顆，即使是相異營運商的衛星，同樣以本地驗證完成換手。 |
| 證據與範圍 | 1,175 · 1,401 · 16.4 · 41.9 · 34.8 ms | 密碼運算的成本可以承受：單一 CPU 核心每秒約可產生 1,175 個、驗證 1,401 個權杖，而在研究的三個星座上花用一個權杖最多需要 16.4、41.9 或 34.8 ms[59, §7.3]。 |
Original figure note原始圖說出處
Drawn for this primer from Liu et al., USENIX NSDI 2024, §3–7, Figs. 7, 15, 17, 19–21.4. Evaluation setup4. 實驗設定
The ground prototype tests commercial phones and SIMs, while trace-driven emulation supplies orbital dynamics. ISL latency, setup and detailed access scheduling remain open modeling inputs.
地面原型測試商用手機及 SIM，軌跡驅動模擬則提供軌道動態，ISL 延遲、建鏈與細部接入排程仍待加入模型。
| Item項目 | Values數值 | Conditions and source條件與來源 |
|---|---|---|
| Evidence and scope | Qualitative scope | The evaluation combines a hardware prototype with data-driven what-if emulation, with orbital flight testing left open [59, §7]. |
| Evidence and scope | 4 · 3 · 50 · 60 | An Amarisoft Callbox NR-4-U runs the satellite’s 3GPP NTN radio stack. Programmable sysmoISIM-SJA2 SIMs store tokens and enforce their use. Huawei Mate 50 and Mate 60 Pro phones provide commercial user devices. The prototype demonstrates end-to-end operation over an NTN air interface, with emulated orbital dynamics. |
| Dataset and workload | Qualitative scope | The emulation replays real orbits from the TLE corpus for Starlink, Globalstar and Iridium. The serviceable-area analysis uses the Starlink Phase II constellation, although the paper leaves unspecified a single planes-by-satellites configuration for the other experiments. |
| Model and compute | 3 | Baselines include 3GPP NR-NTN, IoT-NTN, SpaceCore, and transparent bent-pipe relays. They also include a Starlink model with an onboard radio distributed unit. Cell-mapping comparisons use H3 and latitude–longitude mappings. |
| Latency comparison | 20 · 15 | The design supports satellites with ISLs and satellites with ground links alone. It cites a typical ISL capacity of 20 Gbps [59, §3.1]. ISL latency, topology, link setup, and Starlink’s 15-second reconfiguration require further modeling. |
| 證據與範圍 | 定性範圍 | 實驗結合了硬體原型與資料驅動的假設情境模擬，在軌執行測試則待完成[59, §7]。 |
| 證據與範圍 | 3 · 4 · 50 · 60 | 原型以執行 3GPP NTN 協定堆疊的 Amarisoft Callbox NR-4-U 作為衛星的射頻，以可程式化的 sysmoISIM-SJA2 SIM 卡保存並強制權杖規則，並以 Huawei Mate 50 與 Mate 60 Pro 作為一般商用使用者裝置，所以它證明協定能在商用手機上透過 NTN 空中介面端到端運作，而軌道動態則由模擬提供。 |
| 資料集與工作負載 | 定性範圍 | 模擬重播 TLE 資料集中 Starlink、Globalstar 與 Iridium 的真實軌道，可服務面積分析使用 Starlink Phase II 星座，其他實驗的單一軌道面乘衛星數配置則待指定。 |
| 證據與範圍 | 3 | 它的基準包括 3GPP NR-NTN 與 IoT-NTN、把射頻分散單元放在星上的 Starlink 模型、SpaceCore、透明彎管轉送，以及 H3 或經緯度的服務格對應。 |
| 容量與速率 | 20 · 15 | 論文支援具備 ISL 與僅有地面鏈路的兩種衛星，並引用 20 Gbps 的典型 ISL 容量[59, §3.1]，但它ISL 延遲、拓樸、鏈路建立時間與 Starlink 的 15 秒重新配置都列於模型範圍之外。 |
| Question問題 | Calculation算式 | Result結果 |
|---|---|---|
| Signalling-storm load on the paper's typical ISL論文典型 ISL 上的信令風暴負載 | \(0.1575 \times 20\ \text{Gbps}\) | ≈3.2 Gbpscomputed. 15.75% from Fig. 7, 20 Gbps from §3.1計算值。15.75% 來自 Fig. 7，20 Gbps 來自 §3.1 |
| Same load as a share of a 100 Gbps sustained laser同一負載佔 100 Gbps 持續雷射的比例 | \(3.15 / 100\) | ≈3%computed. 100 Gbps is a company claim via journalists (Chapter 2)計算值。100 Gbps 為經記者轉述的公司說法（第 2 章） |
| Serviceable area relative to baseline相對於基準的可服務面積 | \(1 + 1.16\) | up to 2.16×computed. +116% from §7.2, Fig. 17, Starlink Phase II計算值。+116% 來自 §7.2, Fig. 17，Starlink Phase II |
| Direct-to-cell spectral ceiling for one 5 MHz carrier單一 5 MHz 載波的直連手機頻譜上限 | \(5\ \text{MHz} \times 0.64\ \text{bit/s/Hz}\) | ≈3.2 Mbpscomputed. Median efficiency from [58]計算值。中位數效率來自 [58] |
5. What was measured5. 量到了什麼
| Claim宣稱 | Setup and caveat設定與附註 |
|---|---|
| Serviceable area up to +116% [59, §7.2, Fig. 17]可服務面積最多 +116%[59, §7.2, Fig. 17] | Emulated on Starlink Phase II. "up to" marks the best case across operator mixes.在 Starlink Phase II 上模擬。「最多」表示各營運商組合中的最佳情況。 |
| Service setup up to 5.19× (Starlink), 1.33× (Globalstar), 2.33× (Iridium) faster than NTN [59, §7.2, Fig. 15]服務建立比 NTN 快最多 5.19 倍（Starlink）、1.33 倍（Globalstar）、2.33 倍（Iridium）[59, §7.2, Fig. 15] | Relative figures from emulation. An absolute direct-to-cell latency baseline awaits measurement.來自模擬的相對數字。絕對直連手機延遲的量測錨點仍待建立。 |
| Inter-operator handover signalling 850–7,640× lower, with 4.71–14.25× lower latency [59, §7.4, Fig. 21]跨營運商換手信令低 850–7,640 倍，延遲低 4.71–14.25 倍[59, §7.4, Fig. 21] | The largest ratios reflect removing home-core round trips entirely, so they scale with the baseline's chattiness.最大的比例反映的是完全移除與本籍核心的往返，所以它們隨基準設計的信令頻繁程度而變。 |
| Paging load 23× / 4× / 8× lower. Cell-mapping CPU 100× / 63× / 73× lower [59, Figs. 16, 18]傳呼負載低 23 / 4 / 8 倍。服務格對應 CPU 低 100 / 63 / 73 倍[59, Figs. 16, 18] | Starlink / Globalstar / Iridium. A consequence of Earth-fixed cells.依序為 Starlink / Globalstar / Iridium。是地表固定服務格的結果。 |
| Token spend ≤ 16.4 / 41.9 / 34.8 ms. ≈1,175 generated and 1,401 verified per second per core [59, §7.3]權杖花用 ≤ 16.4 / 41.9 / 34.8 ms。每核心每秒產生 ≈1,175 個、驗證 1,401 個[59, §7.3] | Measured on the prototype hardware, with space-qualified processors left for later testing.在原型硬體上量測，，太空等級處理器則待另行評估。 |
| Worst-case misuse window 0.2–0.88 orbital periods, 22.8–87.7 min [59, Fig. 20]最壞情況的濫用窗口為 0.2–0.88 個軌道週期，22.8–87.7 分鐘[59, Fig. 20] | Applies with ground connectivity only. With ISLs reuse could be reported sooner, at a cost the paper leaves unquantified.適用於僅有地面鏈路時。有 ISL 時可更快回報重用，代價仍待量化。 |
6. Limits and open questions6. 範圍與待解問題
- ISL model scope. The paper treats ISLs as optional and uses a typical 20 Gbps capacity. SpaceX engineers claim 100 Gbps sustained per laser company. At that capacity, the modeled 15.75% signaling share would be about 3%. Per-hop latency, topology, and modern terminals’ 2–30 s setup time require further modeling for ISL-carried reports and signaling.ISL 延遲、拓樸與建立時間模型仍待建立。論文把 ISL 視為可選，並引用「典型」的 20 Gbps，低於 SpaceX 工程師聲稱的每條 Starlink 雷射 100 Gbps 持續速率，所以其 15.75% 的信令比例在這樣的鏈路上約為 3%。每跳延遲、拓樸與現代雷射終端所需的 2–30 s 建立時間都列於模型範圍之外，這對任何經由 ISL 傳送權杖回報或信令的變體都很重要。company
- The result comes from ground-based emulation. The prototype exercises the protocol on phones and SIMs, but the satellite is a ground callbox. The large speedups are relative figures from what-if emulation. Since a measured direct-to-cell latency baseline remains open in the surveyed literature, an absolute latency baseline remains necessary to check the 5.19× setup gain.結果來自地面模擬。原型在手機與 SIM 卡上演練協定，但衛星是一台地面測試設備，而那些大幅加速是假設情境模擬得到的相對數字。本教材所調查文獻中的實測直連手機延遲仍待補充，5.19 倍的建立加速仍需絕對延遲基準才能檢驗。
- The 15-second schedule is absent. Starlink reassigns broadband users globally every 15 seconds, and whether its direct-to-cell payload follows a similar cadence is unknown. Thus, the handover counts in Fig. 21 may understate how often a real Starlink phone is moved between satellites.15 秒排程列於模型範圍之外。Starlink 每 15 秒在全域重新指派寬頻使用者，而其直連手機酬載遵循類似節奏的情況仍待確認，所以 Fig. 21 的換手次數可能低估了真實 Starlink 手機在衛星之間被轉移的頻率。
- Tokens consume uplink capacity. The authors acknowledge the UE-to-satellite bandwidth used by in-band tokens [59, §8]. With median SINR near 0 dB and about 3.2 Mbps on a 5 MHz carrier, overhead per session needs explicit measurement.權杖花用稀缺的上行。作者承認頻內權杖會佔用部分 UE 到衛星的頻寬[59, §8]，而在中位數 SINR 約 0 dB、5 MHz 載波約只有 3.2 Mbps 的鏈路上，每次會談的開銷仍需量化。
- Operator governance needs a deployment model. The paper identifies open questions about verifiable carrier-grade guarantees [59, §8]. Satellite operators control onboard functions, while mobile operators need management arrangements for those functions.治理問題待解。權杖的可驗證電信等級保證仍待建立，而衛星營運商掌管的在軌功能需要與行動營運商建立管理協議[59, §8]，這些正是部署時會遇到的商業障礙。
7. How it connects7. 與其他內容的關聯
- Section 3.6 gives the direct-to-cell spectrum allocation of 2 × 5 MHz and roughly 4 Mbps per beam. These values constrain MOSAIC’s access design.第 3.6 節：該章的直連手機一節給出 2 × 5 MHz 的頻譜與每波束約 4 Mbps，這是 MOSAIC 架構必須在其中運作的條件。
- Section 3.3 describes broadband’s gateway bent pipe. MOSAIC examines the tighter cellular-radio timing requirement, whose scheduling deadline is about 250 µs.第 3.3 節：把寬頻流量送到閘道的彎管，就是 MOSAIC 因 250 µs 期限而為蜂巢射頻所拒絕的「透明管道」。
- Section 5.4: Earth-fixed cells here are the access-layer version of SHORT's geographic routing hierarchy and TinyLEO's geographic anycast.第 5.4 節：這裡的地表固定服務格，是 SHORT 地理路由階層與 TinyLEO 地理任播在存取層的版本。
- Section 13.6 examines access across ground and orbit. MOSAIC provides local satellite admission for cellular users.第 13.6 節：跨地面與軌道服務請求，需要一種讓衛星能在本地接納使用者的存取模型，而 MOSAIC 正為蜂巢使用者提供了這種模型。
- Related digests connect these choices: mobicom24-short and sigcomm25-tinyleo use Earth-fixed geography for routing. pimrc23-pan describes broadband access, while infocom23-ma measures motion effects on dish-based links.相關導讀：mobicom24-short 與 sigcomm25-tinyleo 把地表固定地理用於路由，pimrc23-pan 描述 MOSAIC 所對照的寬頻存取路徑，而 infocom23-ma 量測移動對終端天線鏈路的影響。
Key numbers關鍵數字
| Number | 數字 | Meaning (setup) | 意義（設定） |
|---|---|---|---|
| +116% | Serviceable area, best case, Starlink Phase II emulation [59, §7.2, Fig. 17] | 可服務面積，最佳情況，Starlink Phase II 模擬[59, §7.2, Fig. 17] | |
| 5.19× | Faster service setup than NTN on Starlink. 1.33× Globalstar, 2.33× Iridium [59, Fig. 15] | 在 Starlink 上服務建立比 NTN 快。Globalstar 1.33 倍、Iridium 2.33 倍[59, Fig. 15] | |
| 850–7,640× | Lower inter-operator handover signalling. Latency 4.71–14.25× lower [59, Fig. 21] | 跨營運商換手信令降低倍數。延遲低 4.71–14.25 倍[59, Fig. 21] | |
| 15.75% · 20 Gbps | Peak share of ISL bandwidth consumed by baseline signalling storms · ISL capacity the paper calls typical [59, §3.1–3.2, Fig. 7] | 基準設計信令風暴佔用 ISL 頻寬的峰值比例 · 論文所稱的典型 ISL 容量[59, §3.1–3.2, Fig. 7] | |
| ≈250 µs | Radio deadline a transparent relay to the ground requires a shorter path to meet [59, §3] | 短於透明轉送到地面往返時間的射頻期限[59, §3] | |
| ≤ 16.4 / 41.9 / 34.8 ms | Token spend latency, Starlink / Globalstar / Iridium [59] | 權杖花用延遲，Starlink / Globalstar / Iridium[59] | |
| 22.8–87.7 min | Worst-case token-misuse window with ground connectivity only, 0.2–0.88 orbital periods [59, Fig. 20] | 僅有地面鏈路時最壞情況的權杖濫用窗口，0.2–0.88 個軌道週期[59, Fig. 20] | |
| 50.9 M · 943,003 · 50 | TLE records · Iridium RF measurements · RIPE Atlas Starlink probes in the empirical base [59, Fig. 3] | 實證基礎中的 TLE 紀錄 · Iridium 射頻量測 · RIPE Atlas Starlink 探針[59, Fig. 3] |