Digest · Link physics · read after Chapter 2
導讀 · 鏈路物理 · 第 2 章後閱讀
Laser Inter-Satellite Link Setup Delay: Quantification, Impact, and Tolerable Value雷射星間鏈路建立延遲：量化、影響與可容忍值
Laser-link setup combines pointing and acquisition time. The model charges it once per path change because new links acquire in parallel. A one-second setup raises the example path's average latency substantially, while longer routes tolerate more setup time.雷射鏈路的建立延遲由指向與擷取時間組成。模型假設新鏈路平行擷取，因此每次路徑改變僅計一次建立延遲。一秒的建立時間便會大幅提高範例路徑的平均延遲，而較長路線能容忍較多建立時間。
If you read nothing else
只讀這一段也可以
Every topology and routing paper before this one treated a laser inter-satellite link as instantly available. This preprint is the first to put a price on that assumption. Those values are two to three orders of magnitude below the terminals that fly. These thresholds explain the value of long-lived links. Each dynamic-topology result in the primer requires a comparison between acquisition cost and path savings.在這篇之前的每一篇拓樸與路由論文都把雷射星間鏈路當成可以立即取用，而這篇預印本第一次為那個假設標上價格。這些值比實際飛行的終端低兩到三個數量級，這就是真實星座維持長壽命鏈路的量化理由，也是本教材中每一個動態拓樸結果都必須帶著這個代價來閱讀的原因。
| Measure or comparison參數或比較項目 | Values, conditions and evidence數值、條件與證據 |
|---|---|
| Link setup | Setup delay combines pointing and acquisition time. Terminals need about 2 s for reacquisition with shared ephemeris and 30 s for first acquisition. The STK model uses the 24 × 66 Starlink Phase I shell. Dijkstra routes update every second, with vacuum propagation and 1 ms node delay per hop. Each path change incurs one setup delay. [32, §II–III] |
| Latency comparison | New York–Istanbul uses a 1,700 km laser range. Hour-averaged latency is 25.9 ms (about 26 ms) with zero setup delay. It rises to 35.7 ms at 100 ms setup and 123.9 ms (about 124 ms) at 1 s. Thus, a one-second terminal makes the satellite path slower than fiber by a wide margin [32, Figs. 2–4] . |
| Link setup | Tolerable setup delay is the value at which satellite mean latency equals fiber mean latency. New York–London permits 2.3 ms at 1,500 km range. New York–Hanoi permits 80.63 ms at 5,016 km. [32, Eq. 5, Fig. 4] |
| 鏈路建立 | 作者把建立延遲定義為指向時間加擷取時間，指出目前的終端從有共享星曆的再擷取 2 秒到首次建鏈 30 秒各異，然後在 24 × 66 的 Starlink 第一期軌道殼上，每當 Dijkstra 路徑改變一次就計入這個延遲一次，模擬在 STK 中進行、每秒重算路由、採真空傳播與每跳 1 ms 節點延遲[32, §II–III]。 |
| 延遲比較 | 在紐約–伊斯坦堡、雷射距離 1,700 km 的情況下，一小時平均延遲在建立延遲為零時為 25.9 ms、100 ms 建立延遲時為 35.7 ms、1 秒時為 123.9 ms，所以一秒級的終端會讓衛星路徑比光纖慢上一大截[32, Figs. 2–4]。 |
| 星座配置 | 論文接著把問題反過來，計算衛星路徑平均延遲等於光纖時的可容忍建立延遲，結果是紐約–倫敦在 1,500 km 距離下為 2.3 ms、紐約–河內在 5,016 km 下為 80.63 ms[32, Eq. 5, Fig. 4]。 |
1. Where this paper sits1. 這篇論文的位置
The paper completes a trio from the same Carleton group. The authors list WCNC 2023 as the venue. The fact base verifies only arXiv, so the primer retains the preprint tag. Setup delay is the independent variable in the model.
The number of laser terminals per satellite is absent from the record in the fact base for this paper. Thus, the primer leaves that value open. The companion papers from the same model use four. A reader who needs the figure should check the preprint's system model directly.
這篇論文完成了同一個 Carleton 團隊的三部曲。作者將它列為 WCNC 2023 論文，但事實庫只能驗證 arXiv 版本，所以本教材把它標為預印本。建立延遲是模型的自變數。這篇論文每顆衛星的雷射終端數仍待事實庫補充，因此本教材保留此數值。
同一模型的姊妹論文用四個，需要這個數字的讀者應直接查閱預印本的系統模型。
| Measure or comparison參數或比較項目 | Values, conditions and evidence數值、條件與證據 |
|---|---|
| Link setup | The 2021 classification finds most potential laser neighbors temporary. The 2022 journal paper estimates their benefit under free setup. This preprint asks how much setup can cost before the whole free-space optical network stops beating fiber. |
| Evaluation conditions | Its constellation snapshot is again the 24 × 66, 550 km shell of the November 2018 filing. |
| Delay comparison | Its node-delay assumption is 1 ms per hop, a tenth of the 10 ms the same group used a year earlier. Its terrestrial comparison uses light in fiber at 204,287,876 m/s. This is a refractive index of 1.4675 [32, §II]. |
| 研究問題的承接 | 2021 年的分類顯示一顆衛星大多數潛在雷射鄰居都是暫時的，2022 年的期刊論文顯示若建立時間為零，那些暫時鏈路值多少，而這篇預印本問的是建立時間可以貴到什麼程度，整個自由空間光學網路才會失去相對光纖的優勢。 |
| 評估條件 | 它的星座版本仍是 2018 年 11 月申報的 24 × 66、550 km 軌道殼。 |
| 節點延遲 | 它的節點延遲假設是每跳 1 ms，是同一團隊一年前所用 10 ms 的十分之一，而它的地面比較採用光在光纖中 204,287,876 m/s 的速度，即折射率 1.4675[32, §II]。 |
2. The problem, measured2. 問題，用量測說話
The measurement is a sweep of one parameter through a simulated network. Each path change incurs one delay. All newly required links acquire in parallel, so their attempts count together. The packet that triggers the change waits for the slowest of them.
Dijkstra runs every second for an hour. Each slot with a changed path adds one setup delay to that slot's latency. The hour's average is what the paper reports.
這項量測是在模擬網路中掃描一個參數。塑造這篇論文的建模決定是：延遲每次路徑改變計一次，新鏈路的平行擷取則合併計算，因為最短路徑改變時，它需要而所有需要新建的鏈路會同時建立，觸發改變的封包等待其中最慢的一條。一小時內每秒重跑 Dijkstra，路徑與前一秒相異的每個時槽就把一個建立延遲加到該時槽的延遲上，論文報告的是這一小時的平均。
| Measure or comparison參數或比較項目 | Values, conditions and evidence數值、條件與證據 |
|---|---|
| Link setup | Setup delay runs from a new-pointing command to data exchange, combining pointing and acquisition. Mynaric CONDOR requires about 30 s for first acquisition. After exchanging orbital parameters, later acquisitions require about 2 s. Tesat and General Atomics terminals take tens of seconds. [32, §II] |
| Link setup | At 1,700 km range, New York–Istanbul averages 25.9 ms with free setup. Average latency is 35.7 ms at 100 ms setup and 123.9 ms at 1 s. [32, Figs. 2–4] |
| Link setup | The 9.8 ms and 98 ms increments scale directly with setup delay. They imply roughly one path change per ten seconds over the hour. Geometry and one-second rerouting determine this rate. |
| 鏈路建立 | 建立延遲是指向時間加擷取時間，也就是從終端被告知指向新對象到兩個終端交換資料之間的間隔，作者收集了廠商數字來錨定它：Mynaric CONDOR 首次建鏈約需 30 秒，交換軌道參數後的後續鏈路約 2 秒，Tesat 與 General Atomics 的終端則被描述為需要數十秒[32, §II]。 |
| 鏈路建立 | 對雷射距離 1,700 km 的紐約–伊斯坦堡，這個平均在建立延遲為零時是 25.9 ms、100 ms 時是 35.7 ms、1 秒時是 123.9 ms[32, Figs. 2–4]。 |
| 鏈路建立 | 增量各自為 9.8 ms 與 98 ms，與建立延遲恰成正比，因此隱含路徑在這一小時內以大約每十秒一次的穩定速率改變，這個速率由幾何與每秒重算路由決定，終端則決定每次建立所花的時間。 |
Original figure note原始圖說出處
Drawn for this primer from Bhattacharjee et al., arXiv 2301.05285, Secs. II–III, Eq. 5 and Figs. 2–4.3. The design3. 設計
The construction is easy to follow once the per-path-change rule is accepted. Satellite mean latency combines propagation, node delay and setup delay weighted by the fraction of changed-path slots. Longer laser range reduces hop count and path-change frequency. Both effects reduce mean latency. Fibre latency is fixed by geography. The tolerable delay is therefore the fiber margin divided by the path-change rate, and both numerator and denominator favor long routes and long laser ranges.一旦接受每次路徑改變計一次的規則，這個構造就很容易理解。衛星路徑的平均延遲是它的傳播與節點延遲（隨雷射距離增長而下降，因為更長的鏈路意味更少的跳數），加上建立延遲乘以路徑改變的時槽比例（同樣隨距離下降，因為更稀疏的跳序列較少改變）。光纖延遲由地理固定。因此可容忍延遲就是光纖餘裕除以路徑改變率，分子與分母都偏好長路線與長雷射距離。
| Measure or comparison參數或比較項目 | Values, conditions and evidence數值、條件與證據 |
|---|---|
| Link setup | The paper's design contribution is a metric rather than a mechanism. Equation 5 defines the maximum tolerable setup delay by equality between city-pair hour-averaged satellite latency and fiber latency. Fiber light speed is 204,287,876 m/s. [32, §III] |
| Setup-delay thresholds | The paper's two anchoring cases show the spread: New York–London at a 1,500 km range tolerates only 2.3 ms. The transatlantic satellite path has little propagation margin over fiber. New York–Hanoi at 5,016 km range instead permits 80.63 ms for setup. This occurs because the satellite path saves tens of milliseconds over fiber and changes rarely [32, Fig. 4] . |
| Link setup | Current terminals need 2–30 s. Setup delay must fall by two to four orders of magnitude before dynamic re-pointing pays off in these cases. The results therefore favor links that persist for an orbit. |
| 鏈路建立 | 這篇論文的設計貢獻是一項可量化的評估準則，這個度量是最大可容忍建立延遲，Eq. 5 把它定義為使自由空間光學網路對某組城市對的一小時平均延遲等於同兩城市間地面光纖路徑延遲的建立延遲值，光在光纖中以 204,287,876 m/s 行進[32, §III]。 |
| 星座配置 | 論文的兩個錨定案例顯示出範圍：紐約–倫敦在 1,500 km 距離下只能容忍 2.3 ms，因為跨大西洋光纖已經接近衛星路徑的傳播時間、僅餘極小的建立時間預算，而紐約–河內在 5,016 km 的可見極限下能容忍 80.63 ms，因為衛星路徑比光纖省下數十毫秒而且很少改變[32, Fig. 4]。 |
| 鏈路建立 | 把這些數字對照目前終端的 2–30 秒，就得到論文的結論：建立延遲必須下降兩到四個數量級，動態重新指向的拓樸才划得來，而在那之前，只有能持續一圈軌道的鏈路才值得接線。 |
| Question問題 | Formula公式 | Result結果 |
|---|---|---|
| Per-kilometer delay in fiber at the paper's speed, and the vacuum advantage以論文光纖速度計算的每公里延遲，以及真空的優勢 | \( 1/v_f,\quad v_f/c - 1 \) | 4.895 µs/km, 46.75 %computed. V_f = 204,287,876 m/s (n = 1.4675, Sec. II). Vacuum 3.336 µs/km計算值。v_f = 204,287,876 m/s（n = 1.4675，Sec. II）。真空 3.336 µs/km |
| Penalty per second of setup delay, from the two non-zero cases由兩個非零案例算出的每秒建立延遲代價 | \( (35.7-25.9)/0.1,\;(123.9-25.9)/1 \) | 98 ms per s, bothcomputed from Figs. 2–4 values. The match shows the penalty is linear in setup delay由 Figs. 2–4 的數值計算。兩者相符顯示代價與建立延遲呈線性 |
| Implied path-change rate, if the penalty equals setup delay times change rate as the once-per-path-change rule implies若代價等於建立延遲乘以改變率（如每次路徑改變計一次的規則所隱含），隱含的路徑改變率 | \( r = 0.098\ \text{s}^{-1} \) | ≈ 1 change per 10.2 scomputed. Our inference from the paper's numbers, New York–Istanbul, 1,700 km, Dijkstra every 1 s計算值。我們從論文數字的推論，紐約–伊斯坦堡、1,700 km、每秒 Dijkstra |
| Spread of tolerable setup delay between the short and the long route短路線與長路線可容忍建立延遲的差距 | \( 80.63/2.3 \) | 35×computed from Fig. 4 values由 Fig. 4 數值計算 |
| How far current terminals miss the tolerable values目前終端距可容忍值差多少 | \( 2\ \text{s}/80.63\ \text{ms},\; 2\ \text{s}/2.3\ \text{ms} \) | 25×, 870×computed. 2 s is the CONDOR re-acquisition figure relayed in Sec. II. A 30 s first link is 15× worse again計算值。2 秒是 Sec. II 轉述的 CONDOR 再擷取數字。30 秒的首次建鏈再差 15 倍 |
| One-way propagation over one hop at the two ranges used for the tolerable values兩個可容忍值所用距離下單跳的單向傳播 | \( d/c \) | 5.0 ms, 16.7 mscomputed. 1,500 km and 5,016 km計算值。1,500 km 與 5,016 km |
4. Evaluation setup4. 實驗設定
The study is a simulation. The constellation snapshot is the November 2018 filing. This is superseded. The node-delay assumption is 1 ms per hop.
Fibre is modeled as light at 204,287,876 m/s along the terrestrial route. Thus, the comparison is propagation-only on both sides and ignores the measured Internet's extra routing and queuing. Capacity, traffic and link rate lie outside the model.
這項研究是模擬。節點延遲假設為每跳 1 ms。光纖被建模為沿地面路線以 204,287,876 m/s 行進的光，所以比較兩側都只算傳播，忽略實測網際網路額外的路由與排隊。容量、流量與鏈路速率皆位於模型範圍之外。
| Measure or comparison參數或比較項目 | Values, conditions and evidence數值、條件與證據 |
|---|---|
| Link setup | STK propagates the 24 × 66 Starlink Phase I shell at 550 km, Dijkstra is run every second over an hour. Each path change adds one setup delay to the latency of that slot [32, §II–III]. |
| Link setup | Setup delay is modeled and swept, with 100 ms and 1 s as the illustrated values and the tolerable value solved for in Fig. 4. |
| Delay comparison | The New York–Istanbul curves use 1,700 km laser range. The two tolerable-delay examples use 1,500 and 5,016 km. |
| Link setup | The number of terminals per satellite is absent from the record in the fact base for this preprint. The vendor setup figures of 30 s and 2 s are second-hand, taken from Mynaric material through the paper rather than from a datasheet. |
| 鏈路建立 | STK 推演 550 km 的 24 × 66 Starlink 第一期軌道殼，一小時內每秒跑一次 Dijkstra，每次路徑改變就把一個建立延遲加到該時槽的延遲上[32, §II–III]。 |
| 評估條件 | 星座版本是已被取代的 2018 年 11 月申報。 |
| 鏈路建立 | 建立延遲有建模並被掃描，以 100 ms 與 1 秒為圖示值，可容忍值則在 Fig. 4 中求解。 |
| 延遲比較 | 雷射距離是參數，紐約–伊斯坦堡的代價曲線用 1,700 km，兩個可容忍延遲錨點用 1,500 與 5,016 km。 |
| 終端與建立時間來源 | 這篇預印本每顆衛星的終端數在仍待事實庫補充，而 30 秒與 2 秒的廠商建立數字是二手的，透過論文取自 Mynaric 的資料規格書證據則仍待補充。 |
5. What was measured5. 量到了什麼
| Claim宣稱 | Setup and caveat設定與附註 |
|---|---|
| Current terminals need 2–30 s to set up a link (CONDOR 30 s first, 2 s later. Tesat and General Atomics tens of seconds).目前終端需要 2–30 秒建立鏈路（CONDOR 首次 30 秒、之後 2 秒。Tesat 與 General Atomics 數十秒）。 | Sec. II, vendor figures cited second-hand. A direct datasheet source remains to be established, and the SDA standard separately allows ≤100 s cold start.Sec. II，二手引用的廠商數字。直接提供數值的規格書仍待補充，SDA 標準另允許冷啟動 ≤100 秒。 |
| New York–Istanbul at 1,700 km: mean latency is 25.9 ms with zero setup. It reaches 35.7 ms at 100 ms setup and 123.9 ms at 1 s.紐約–伊斯坦堡、1,700 km：建立延遲為零時平均 25.9 ms，100 ms 時 35.7 ms，1 秒時 123.9 ms。 | Figs. 2–4 and text. One hour, Dijkstra every second, 1 ms node delay, 24 × 66 shell. One-way. The penalty is linear in setup delay.Figs. 2–4 與內文。一小時、每秒 Dijkstra、1 ms 節點延遲、24 × 66 軌道殼。單向。代價與建立延遲呈線性。 |
| New York–London permits 2.3 ms setup at 1,500 km range. New York–Hanoi permits 80.63 ms at 5,016 km.可容忍建立延遲在紐約–倫敦、1,500 km 下為 2.3 ms，在紐約–河內、5,016 km 下為 80.63 ms。 | Eq. 5 and Fig. 4. Equality with propagation-only fiber at n = 1.4675. Against the measured Internet, which runs slower than fiber propagation, the tolerable values would be larger.Eq. 5 與 Fig. 4。與僅計傳播、n = 1.4675 的光纖相等。若對照比光纖傳播更慢的實測網際網路，可容忍值會更大。 |
| Setup delay is charged once per path change, with parallel new-link acquisition counted together.建立延遲每次路徑改變計一次，新鏈路的平行擷取則合併計算。 | Sec. II modeling rule. Assumes all new links on a path acquire in parallel and that routing reroutes at every shortest-path change.Sec. II 的建模規則。假設一條路徑上所有新鏈路同時擷取，且路由在每次最短路徑改變時都重新選路。 |
6. Limits and open questions6. 限制與待解的問題
- Preprint status and a superseded shell. The WCNC 2023 listing awaits verification. Thus, every number here carries a preprint tag. The 24 × 66 shell was replaced by 72 × 22 in August 2019 (fact base Part C, item 12). The flown intra-plane hop is 1,969 km. The 1,500 and 1,700 km cases therefore fall short of that neighbor distance. Their path-change rates would differ.預印本身分與已被取代的軌道殼。WCNC 2023 的列名仍待驗證，所以此處每個數字都帶預印本標籤，而 24 × 66 軌道殼在 2019 年 8 月被 72 × 22 取代（事實庫 Part C 第 12 項）。在實際飛行的幾何中同軌道面跳距為 1,969 km，所以 1,500 與 1,700 km 的距離情況低於面內鄰居所需的距離，路徑改變率也會各異。
- The penalty is for the most naive router. Recomputing Dijkstra every second and switching at every shortest-path change maximizes path changes. A router that holds a path until its links break, or TinyLEO's matching that favors long-lived links [39, §4], would pay setup delay far less often. Thus, the 98 ms-per-second penalty is an upper bound for the modeled geometry rather than a property of laser links.代價是針對最天真的路由器。每秒重算 Dijkstra 並在每次最短路徑改變時切換，會使路徑改變次數最大化。一個維持路徑直到其鏈路斷掉的路由器，或是 TinyLEO 偏好長壽命鏈路的配對[39, §4]，付建立延遲的次數會少得多，所以每秒 98 ms 的代價是對所建模幾何的上界，其適用範圍限於這套幾何與路由規則。
- Vendor figures are second-hand and the CONDOR rate claim requires evidence beyond the product page. The paper relays 30 s and 2 s CONDOR values from Mynaric material. Direct datasheet verification remains pending (Part C, item 18 company ). The current CONDOR page lists 0.313–2.5 Gbps, with 10 Gbps upcoming (Part D, item 4). The associated "up to 100 Gbps" figure requires another source. Setup time and rate should both be treated as vendor claims of their date.廠商數字是二手的，且 CONDOR 的速率宣稱仍待產品頁以外的證據。CONDOR 的 30 秒與 2 秒數值是透過這篇論文對 Mynaric 資料的引用進入本教材，規格書證據則仍待補充（Part C 第 18 項 company），而與之一同流傳的「最高 100 Gbps」CONDOR 數字仍待目前產品頁以外的證據，該頁列的是 0.313–2.5 Gbps、10 Gbps 即將推出（Part D 第 4 項）。建立時間與速率都應視為帶有日期的廠商宣稱。
- Node delay is 1 ms here and 10 ms in the companion paper. Both remain unmeasured (Part C, item 19). Node delay shifts the zero-setup baseline and fiber margin. Both 2.3 and 80.63 ms thresholds would change under the 10 ms assumption. This makes a long path's baseline tens of milliseconds slower and its margin over fiber smaller.這裡的節點延遲是 1 ms，姊妹論文是 10 ms。兩者都量測資料仍待補充（Part C 第 19 項）。這個選擇會移動建立延遲為零的基準，因而移動光纖餘裕，所以 2.3 與 80.63 ms 的可容忍值在 10 ms 假設下都會改變，因為那會使長路徑的基準慢上數十毫秒、對光纖的餘裕變小。
- Fibre is propagation-only. The comparison uses light in glass as a strict baseline. Handley finds measured New York–London Internet latency 21 ms above great-circle fiber. [3, Fig. 7] Against real paths the satellite network would tolerate more setup delay. Equally, the satellite side omits the measured 40 ms bent-pipe floor of Chapter 3. Thus, both sides represent propagation-model estimates.光纖只計傳播。與玻璃中的光相比、實測網際網路則另有路由及排隊因素[3, Fig. 7]，是嚴格的選擇。對照真實路徑，衛星網路能容忍更多建立延遲。同樣地，衛星側省略了第 3 章實測的 40 ms 彎管地板，所以兩側都屬於端到端範圍以外的數字。
- Terminal count and parallel acquisition are unstated here. The fact base leaves the terminals-per-satellite figure open for this preprint. The once-per-path-change rule assumes every new link on the path acquires simultaneously. This requires a free terminal at each end of each new link. With three terminals per current Starlink satellite (Part C, item 13 company) that spare capacity still requires a hardware feasibility check.終端數與平行擷取在此說明資料待補。這篇預印本每顆衛星的終端數仍待事實庫補充，而每次路徑改變計一次的規則假設路徑上每條新鏈路同時擷取，這要求每條新鏈路兩端各有一個空閒終端。以目前 Starlink 衛星每顆三個終端（Part C 第 13 項 company）來看，這種備用容量仍須核對硬體可提供的備用容量。
7. How it connects7. 與其他內容的關聯
- Section 2.1 compares propagation at 3.336 versus 4.895 µs/km, defining the fiber margin available for setup. It also compares 1 ms and 10 ms node delays.第 2.1 節：定義本文花在建立延遲上的光纖餘裕的 3.336 對 4.895 µs/km 比較，以及 1 ms 對 10 ms 的節點延遲選擇。
- Section 2.4: pointing, acquisition and the 2–30 s setup figures, with the ~30 acquisitions per laser per day that SpaceX's own numbers imply.第 2.4 節：指向、擷取與 2–30 秒的建立數字，以及 SpaceX 自身數字隱含的每個雷射每天約 30 次擷取。
- Section 2.5: the tolerable 2.3–80.63 ms range as the quantitative reason topologies are built from persistent links.第 2.5 節：2.3–80.63 ms 的可容忍範圍，作為拓樸由持久鏈路構成的量化理由。
- Section 2.3: the SDA standard's ≤100 s cold-start requirement, the only setup figure from a primary document.第 2.3 節：SDA 標準 ≤100 秒的冷啟動要求，唯一來自一手文件的建立時間數字。
- Section 5.1 and Section 5.7: the fiber-beating argument this paper qualifies, and the table of topology papers that ignore setup delay.第 5.1 節與第 5.7 節：本文加以限定的勝過光纖論證，以及忽略建立延遲的拓樸論文一覽表。
- Sibling digests ojcoms22-temporary-isl and vtm21-chaudhry cover free-setup bounds and link classes. The fiber-comparison framing comes from hotnets18-handley. Link-stable approaches appear in sigcomm25-tinyleo and arxiv26-ali-zhang.同組導讀：ojcoms22-temporary-isl 是本文修正的樂觀上界，vtm21-chaudhry 給出鏈路類型，hotnets18-handley 提供與光纖比較的框架，sigcomm25-tinyleo 展示一個避免路徑改變的控制器，arxiv26-ali-zhang 則只用一圈軌道內穩定的鏈路。
Key numbers關鍵數字
| Number | 數字 | Meaning (setup) | 意義（設定） |
|---|---|---|---|
| 24 × 66, 550 km | Constellation snapshot, November 2018 filing. STK, Dijkstra every 1 s for one hour [32, §II–III] | 星座版本，2018 年 11 月申報。STK、一小時內每秒 Dijkstra[32, §II–III] | |
| 1 ms / hop | Node delay assumed. Unmeasured, and 10× smaller than the companion OJCOMS paper | 假設的節點延遲。量測資料仍待補充，且比姊妹 OJCOMS 論文小 10 倍 | |
| 204,287,876 m/s | Speed of light in fiber used for the terrestrial comparison, n = 1.4675 [32, §II] | 地面比較所用的光纖光速，n = 1.4675[32, §II] | |
| 30 s / 2 s | CONDOR first-link and re-acquisition setup times, cited from vendor material [32, §II] | CONDOR 首次建鏈與再擷取的建立時間，引自廠商資料[32, §II] | |
| 25.9 → 35.7 → 123.9 ms | New York–Istanbul average at 1,700 km with 0, 100 ms and 1 s setup delay [32, Figs. 2–4] | 紐約–伊斯坦堡在 1,700 km 下，建立延遲 0、100 ms 與 1 秒時的平均值[32, Figs. 2–4] | |
| 98 ms per s | Penalty slope, computed from the two non-zero cases. Implies ≈ 0.098 path changes per second | 代價斜率，由兩個非零案例計算。隱含每秒約 0.098 次路徑改變 | |
| 2.3 ms | Tolerable setup delay, New York–London at 1,500 km laser range [32, Eq. 5, Fig. 4] | 可容忍建立延遲，紐約–倫敦、1,500 km 雷射距離[32, Eq. 5, Fig. 4] | |
| 80.63 ms | Tolerable setup delay, New York–Hanoi at the 5,016 km visibility limit [32, Eq. 5, Fig. 4] | 可容忍建立延遲，紐約–河內、5,016 km 可見極限[32, Eq. 5, Fig. 4] | |
| once per path change | Accounting rule for setup delay: parallel new-link acquisitions counted together [32, §II] | 建立延遲的計算規則：各條新鏈路的平行擷取合併計算[32, §II] | |
| absent from the record | Laser terminals per satellite for this preprint, in the fact base. The companion papers use four | 事實庫中這篇預印本每顆衛星的雷射終端數。姊妹論文用四個 |