
Six days in, and today's post is basically the payoff for the whole week — putting LoRaWAN, Meshtastic, and MeshCore side by side and being honest with myself about which one actually fits the problem Siliqs is solving.
LoRaWAN is an infrastructure network. Every end device talks to one or more fixed gateways within radio range, single-hop only — the device itself never relays another device's traffic. The gateway forwards everything up to a network server, usually over normal internet backhaul. This is a genuinely good model when you want centralized management, when devices are spread across a wide area that already has (or can justify) gateway coverage, and especially when you want to build on a standardized, interoperable protocol that lots of vendors' hardware and lots of public/commercial network infrastructure already speaks. The catch, as I understand it now, is that you need gateway coverage everywhere your devices are. No gateway in range, no connectivity — full stop.
Meshtastic and MeshCore are both peer-to-peer mesh networks, which is a structurally different shape. There's no fixed infrastructure requirement at all — every node can potentially relay for other nodes, hopping a message across multiple hops to reach one that's out of direct range of any single node. Coverage isn't "wherever there's a gateway," it's "wherever there's a chain of nodes close enough to hear each other." They differ from each other mainly in routing approach — Meshtastic's flood routing versus whatever MeshCore is doing differently, which I'm still working out from yesterday's post — but structurally, from a topology standpoint, they're the same basic shape: distributed, no central relay point required.
Now, why does this matter for Siliqs specifically? Our core use case is RS485 sensors spread across an industrial site — think a factory floor, a warehouse, a facility with buildings or structures that block direct line of sight. A few things about that scenario point pretty clearly toward the mesh model over LoRaWAN:
- No guaranteed gateway coverage everywhere on-site. Industrial sites have metal structures, concrete, machinery — a lot of RF-unfriendly stuff. Relying on a single gateway (or even a few) covering every sensor directly is a much bigger ask than letting nodes relay for each other and route around obstacles.
- The network doesn't need to talk to the public internet at every point. A private industrial mesh can be entirely self-contained on-site, with only a single point (or a small number) needing to bridge out. LoRaWAN's whole design assumes gateways with backhaul; a mesh doesn't need that at every node.
- Deployment flexibility. Adding a new sensor node in a mesh mostly just means it needs to be in range of some existing node, not specifically a gateway. That's a much lower bar for site coverage as the deployment grows or changes.
I don't want to overstate this into "mesh is strictly better" — LoRaWAN's star topology has real advantages for things like standardization, roaming across operator networks, and simpler network-side logic, and I'm sure there are tradeoffs on the mesh side (routing complexity, and I'd guess harder worst-case latency bounds since a message might need several hops) that I haven't fully explored yet. But for "sensors scattered across one physical site, need to relay past obstacles, don't need public internet infrastructure everywhere" — which is genuinely close to describing what our SQC485Iv2 deployments look like — the mesh model is the more natural fit, and now I understand why rather than just having been told so.
Tomorrow, last day of RF fundamentals week: the actual modulation parameters — spreading factor, bandwidth, coding rate — the knobs that determine range, speed, and robustness on a LoRa link.

進入第六天,今天這篇基本上是這整週的總結——把 LoRaWAN、Meshtastic、MeshCore 放在一起比較,然後老實面對自己:到底哪一個真正適合 Siliqs 要解決的問題。
LoRaWAN 是一個基礎設施型網路。 每個終端裝置都是跟訊號範圍內的一個或多個固定 gateway 通訊,而且只有單一跳(single-hop)——裝置本身從來不會幫其他裝置轉發流量。gateway 把所有東西往上轉送給 network server,通常是透過一般的網際網路 backhaul。當你想要集中管理、裝置分散在一個已經有(或值得建置)gateway 涵蓋範圍的廣大區域,尤其是想建立在一個標準化、可互通的協定之上——很多廠商的硬體跟很多公開/商用網路基礎設施都已經支援——這其實是個很不錯的模型。但照我現在的理解,代價是:你的裝置在哪,就得在哪有 gateway 涵蓋。範圍內沒有 gateway,就完全沒有連線,沒有例外。
Meshtastic 跟 MeshCore 都是點對點的 mesh 網路,結構上是完全不同的形狀。完全不需要固定的基礎設施——每個節點原則上都能幫其他節點轉發,讓一則訊息跳過多個節點去接觸單一節點直接觸及不到的地方。涵蓋範圍不是「哪裡有 gateway」,而是「哪裡有一串距離夠近、能互相聽到的節點」。這兩者之間主要的差別在於 routing 做法——Meshtastic 的 flood routing,對比 MeshCore 不太一樣的做法(我昨天那篇還在搞懂中)——但就拓樸結構來說,它們是同一種基本形狀:分散式、不需要中心轉發點。
那這對 Siliqs 具體來說為什麼重要?我們的核心使用情境是分散在工業現場的 RS485 感測器——想像一下工廠廠房、倉庫、有建築物或結構會擋住直接視線的場地。這個情境有幾件事,蠻明確地指向 mesh 模型會比 LoRaWAN 更合適:
- 現場不是每個角落都保證有 gateway 涵蓋。 工業現場有金屬結構、混凝土、機具——很多對 RF 不友善的東西。指望單一個(甚至幾個)gateway 直接涵蓋每一個感測器,是比讓節點互相幫忙轉發、繞過障礙物大得多的要求。
- 網路不需要每個點都連上公開網際網路。 私有工業 mesh 可以完全侷限在現場內部運作,只需要單一個(或少數幾個)點負責橋接出去就好。LoRaWAN 的整個設計是假設 gateway 都有 backhaul;mesh 則不需要每個節點都有。
- 部署彈性。 在 mesh 裡新增一個感測器節點,大致上只需要它在「某個」既有節點的範圍內就好,不用特別是 gateway。隨著部署規模擴大或改變,這對場地涵蓋率來說是低得多的門檻。
我不想把這件事講成「mesh 絕對比較好」——LoRaWAN 的星狀拓樸在標準化、跨營運商網路漫遊、network 端邏輯比較簡單這些方面都有真正的優勢,而 mesh 這邊我猜也有還沒完全摸清楚的取捨(routing 複雜度,我猜最差情況的延遲上限大概也比較難保證,因為一則訊息可能得跳好幾次)。但對「感測器散布在單一個實體場地、需要繞過障礙物轉發、不需要每個點都有公開網際網路基礎設施」這種情境——這其實蠻貼近我們 SQC485Iv2 實際部署的樣子——mesh 模型是比較自然的選擇,而現在我理解的是「為什麼」,而不只是被告知答案而已。
明天是 RF 基礎週的最後一天:實際的調變參數——spreading factor、bandwidth、coding rate——這些決定一條 LoRa 鏈路的距離、速度跟穩健度的旋鈕。