"Day 30/30: Looking Back at 30 Days — From 'What's LoRaWAN?' to Actually Deploying a Node"

a calendar with 30 marked days

Thirty days ago I mixed up LoRa and LoRaWAN in my very first onboarding conversation with Wei. Today I flashed a node, configured its role and mesh settings, and could explain why it needs a gateway on-site to answer a customer's SCADA system. I want to actually trace that arc instead of just asserting it happened, because writing it out is the only way I can tell whether I really understand this stuff or just got used to the vocabulary.

Weeks 1-2 were about the radio and the mesh. Day 1 was LoRa versus LoRaWAN — the physical layer versus the network protocol, and the realization that Siliqs' firmware uses LoRa radios but doesn't speak LoRaWAN at all, which confused me for longer than I'd like to admit. Day 2 was RF regulations, which explained why region settings exist at all — something that stopped being abstract the moment I actually had to pick a region in the Configurator on day 28. Days 4 through 8 built up mesh networking versus classic LoRaWAN star topology, and days 9 through 12 got into the mechanics — flood routing, channel/PSK privacy, SNR/RSSI, and the SIR-driven collision behavior between spreading factors. Day 13 and 14 were antennas and VNA calibration, which felt like a detour into pure RF measurement at the time but turned directly into day 29's "a flaky link is often a connector problem, not a firmware bug" lesson.

Weeks 2-3 were RS485 and Modbus. Day 15 through 19 built up from RS485's differential signaling all the way to why Modbus RTU, a 1979 protocol, is still the default on industrial sites, and how it rides on top of a LoRa mesh instead of getting replaced by one. This is the part of the series that mattered most once I got to the real product, because the SQC485Iv2's entire job on the wired side is speaking Modbus RTU as a master.

Week 4 was embedded firmware constraints — flash and RAM as genuinely scarce, separately-budgeted resources (day 20), why OTA needs two firmware slots instead of one (day 21), deep sleep and duty cycle as the real levers for battery life (day 22), the ADC/resistor-divider chain for reading battery voltage honestly (day 23), and flash budget eating more engineering time than I expected features to (day 24). At the time these felt like a separate, more "electronics fundamentals" track from the RF and Modbus weeks. They weren't — they're constraints that shape everything the product can actually do.

And then this final week was where it all landed on one real device. Day 25 was the moment the whole picture actually assembled itself — an RS485 Modbus master, reporting over a Meshtastic-based private LoRa mesh, running on firmware with the flash and power budget I'd just spent a week learning to respect. Day 26 was the role split that makes a deployment actually work end to end: field/sensor mode polling a local bus, gateway mode caching mesh-wide readings and answering a downstream Modbus master instantly with a staleness check, plus the transparent-versus-aggregated addressing choice I still want a real example of. Days 27 and 28 were the two tools that turn all of that from "edit firmware source and reflash" or "type exact serial commands" into something a field technician could reasonably do — flasher.siliqs.net and mesh.siliqs.net's Configurator, provision then configure, the first two-thirds of what Siliqs calls "裝 → 設 → 接." Day 29 was stepping back to write down the debugging instincts I've actually started forming on the bench, independent of any one product.

What I'd tell myself on day 1: the confusing part isn't any single layer, it's that a real product is all of these layers stacked on top of each other with real tradeoffs between them, and you can't actually reason about the whole thing until you've spent time in each layer separately first. I don't think I could have skipped any of the four weeks and still understood day 25 the way I did.

What's still genuinely fuzzy for me, and what I want to chase next: I've never seen an actual customer's aggregated-addressing config, so I want to sit with Amy and look at what real deployments actually choose and why. I also want to get back into RF measurement — the VNA and antenna work from week 2 was the part I understood least deeply at the time, and I suspect a proper S11 measurement on a deployed node would teach me things the datasheet alone can't. And honestly, I want to see what happens when something in the field actually breaks — every lesson from day 29 was theoretical today, and I want to find out which of those instincts actually hold up against a real problem.

月曆上標示 30 天

三十天前,我在跟 Wei 第一次 onboarding 對話裡就把 LoRa 跟 LoRaWAN 搞混了。今天,我燒錄了一台 node、設定了它的角色跟 mesh 設定,還能解釋為什麼現場需要一台 gateway 才能回應客戶 SCADA 系統的查詢。我想把這條軌跡真的追一遍,而不只是說「我進步了」——把它寫出來,是我唯一能確認自己是真的搞懂這些東西、還是只是習慣了這些詞彙的方法。

第 1-2 週是無線電跟 mesh。 Day 1 是 LoRa 跟 LoRaWAN 的差別——實體層跟網路協定層的區別,以及意識到 Siliqs 韌體用的是 LoRa radio,但完全不講 LoRaWAN,這件事讓我困惑的時間比我願意承認的還久。Day 2 是 RF 法規,解釋了地區設定為什麼存在——這件事在 day 28 我真的要在 Configurator 裡選一個地區的那一刻,就不再是抽象概念了。Day 4 到 8 建立起 mesh networking 跟傳統 LoRaWAN 星狀拓樸的對比,day 9 到 12 深入機制——flood routing、channel/PSK 隱私、SNR/RSSI,以及不同展頻因子之間由 SIR 主導的碰撞行為。Day 13、14 是天線跟 VNA 校正,當時感覺像是繞去做純 RF 量測的插曲,但後來直接變成 day 29 那句「連線不穩通常是接頭問題,不是韌體 bug」的教訓來源。

第 2-3 週是 RS485 跟 Modbus。 Day 15 到 19 從 RS485 的差動訊號一路建立到「為什麼一個 1979 年的協定 Modbus RTU 到現在還是工業現場的預設選擇」,以及它怎麼搭在 LoRa mesh 上、而不是被 mesh 取代。這段是整個系列裡對我理解真實產品最關鍵的部分,因為 SQC485Iv2 在有線那一側的全部工作,就是扮演 Modbus RTU 的 master。

第 4 週是嵌入式韌體限制——flash 跟 RAM 是真正稀缺、要分開編列預算的資源(day 20)、為什麼 OTA 需要兩個韌體 slot 而不是一個(day 21)、深度睡眠跟 duty cycle 才是真正決定電池壽命的槓桿(day 22)、用 ADC/分壓電阻鏈誠實地讀出電池電壓(day 23)、以及 flash 預算吃掉的工程時間比我以為的還多(day 24)。當時這些感覺像是跟 RF、Modbus 那幾週分開的、比較「電子學基礎」的一條支線。但其實不是——它們是塑造這個產品實際能做到什麼的限制條件。

而這最後一週,就是把一切都落到一台真實裝置上的時刻。 Day 25 是整幅圖真正自己拼起來的那一刻——一台 RS485 Modbus master,透過 Meshtastic-based 的私有 LoRa mesh 回報資料,跑在我才剛花一週學會要尊重的 flash 跟電源預算的韌體上。Day 26 是讓一個部署真正端到端跑得起來的角色區分:field/sensor mode poll 本地 bus,gateway mode 快取整個 mesh 的讀數、並用附帶 staleness check 的方式立即回應下游 Modbus master,再加上我還想找到真實案例的 transparent vs aggregated 定址選擇。Day 27、28 是把上面這一切從「改韌體原始碼再重新燒錄」或「打精確的 serial 指令」變成現場技術人員合理做得到的兩個工具——flasher.siliqs.net 跟 mesh.siliqs.net 的 Configurator,也就是 provision 然後 configure,是 Siliqs 說的「裝 → 設 → 接」的前三分之二。Day 29 是退一步,把我在 bench 上真正養成的除錯直覺寫下來,跟任何單一產品無關。

如果要跟 day 1 的自己講一句話:真正讓人困惑的不是任何單一一層,而是一個真實產品是這些層彼此疊在一起、之間有真實 tradeoff 的整體,你沒辦法在還沒分別花時間搞懂每一層之前,就跳過去推理整體。我不覺得自己能跳過任何一週,還能像今天這樣理解 day 25 的內容。

對我來說目前還真的模糊的地方、也是接下來想追的方向:我還沒看過真實客戶的 aggregated 定址設定長什麼樣子,所以想找 Amy 坐下來,看看真實部署到底都怎麼選、為什麼這樣選。我也想回頭再深入 RF 量測——第 2 週的 VNA 跟天線那部分是我當時理解得最不深的地方,我猜在一台已部署的 node 上做一次正規的 S11 量測,能教我一些光看 datasheet 學不到的東西。老實說,我也很想看看現場真的出問題時會發生什麼——day 29 寫的每一條教訓今天都還只是理論,我想知道這些直覺真的碰到實際問題時,哪些撐得住。