From ../component-selection.md and
../documentation.md §2–4.
Component-selection questions are really “can you justify a decision under pressure?” questions.
Every answer should be requirement → candidates → chosen → what you gave up.
Q1. [Intermediate] 🔥 Why the ESP32-C6 specifically? There are cheaper MCUs.
Strong answer covers: it’s the radio complement that decides it. One chip gives Wi-Fi 6
(802.11ax, 2.4 GHz), Bluetooth 5.0 LE, and an 802.15.4 radio — which is what Zigbee 3.0 and Thread
1.3 run on. For a bridge whose entire job is speaking to heterogeneous local devices, getting
Zigbee/Thread without a second radio IC is the whole argument. Plus a single RISC-V core at 160 MHz,
512 KB SRAM, 8 MB on-module flash (N8), 23 GPIO, SPI/I2C/UART/USB-Serial-JTAG, and −40 to +85 °C.
The constraint to name: Zigbee and Thread share the same 802.15.4 radio and are time-shared,
so only one is active at a time — firmware-selectable per deployment, not simultaneous. Volunteering
that limit is what makes the answer credible.
Q2. [Intermediate] 🔥 Module or bare chip? Defend it.
Strong answer covers: the WROOM-1 module, for three reasons that all reduce risk rather than
cost: the antenna matching network is already designed and tuned (RF layout is the easiest thing to
get wrong on a first board), decoupling is integrated, and the module carries FCC/CE
certification — so the product inherits it instead of paying for full intentional-radiator testing.
The cost: higher unit price, a larger footprint, and no freedom to change the RF front end. The
rule to state: a bare chip is the right call at volume, when the RF layout is a solved problem for
you; a module is right for a first spin where certification and RF risk dominate the BOM delta.
Q3. [Advanced] 🔥 Why 868 MHz for LoRa rather than 915 or 433?
Strong answer covers: Nigeria is ITU Region 1, the same regulatory region as Europe, so
868 MHz is the correct ISM band. 915 MHz is Region 2 (the Americas); using it here would be a
regulatory violation, not just a suboptimal choice. This is a deployment-geography decision, not a
technical preference — and the SX1262 covers 150–960 MHz, so the chip isn’t the constraint, the
antenna, matching network and filter are.
Follow-up worth pre-empting: “What changes if you sell into the US?” — a different antenna and
matching network, a re-tune, and different duty-cycle/power rules. The chip stays.
Q4. [Intermediate] Walk me through the SX1262’s key specs and why they matter for this application.
Strong answer covers, with the why attached to each number:
- +22 dBm TX / −148 dBm RX sensitivity → the link budget, which is what buys the kilometres.
LoRa’s whole proposition is range, and that pairing is where it comes from. - 4.2 mA RX current, <1.5 µA sleep → the node side of the network can run for years on a
battery; the gateway benefits less, but the same radio is used at both ends. - SPI + 3 DIO pins → simple MCU interface, but the DIOs matter: DIO2 can drive the RF switch
automatically (see 02-power-and-rf.md Q6). - LoRa (CSS) plus FSK/GFSK → chirp spread spectrum is what gives sub-noise-floor reception; FSK
is there for compatibility with conventional links.
Q5. [Intermediate] 🔥 Why the W5500 rather than using the ESP32’s own MAC with an external PHY?
Strong answer covers: the W5500 is a hardwired TCP/IP stack — TCP, UDP, IPv4, ICMP, ARP,
IGMP, PPPoE and eight independent hardware sockets, all in silicon. The MCU talks to it over SPI and
never runs a software network stack for Ethernet, which frees both RAM and CPU on a single-core part
that’s already juggling three radios.
The trade-off: SPI throughput caps you well below wire speed, and you’re limited to the protocols
and socket count the chip implements — no custom stack behaviour, no raw-socket tricks. For a
telemetry bridge moving small messages that’s irrelevant; for a high-throughput application it would
be the wrong choice. Also worth naming: an EMAC+PHY design is cheaper in parts and more flexible, and
costs you firmware complexity and RAM.
Q6. [Intermediate] Why the SIM7080G, and what did you give up?
Strong answer covers: it targets exactly the right tier — LTE Cat-M1 and NB-IoT, the
low-power wide-area cellular standards built for telemetry, with EGPRS/2G fallback which still
matters in regions where LTE-M coverage is patchy. Bonus GNSS (GPS/GLONASS/BeiDou) at no extra part
cost. Interface is UART with AT commands — simple to drive, no high-speed bus needed at ~600 kbps.
What you give up: speed (this is not an LTE modem for bulk data), and no onboard eSIM — so
there’s a physical nano-SIM slot, with eSIM pads laid out as DNP (do-not-populate) reserved for a
future revision. That last detail is a good one to volunteer: it’s cheap to reserve a footprint now
and impossible to add later without a respin.
Q7. [Advanced] 🔥 Everything hangs off SPI. How did you decide the bus speed?
Strong answer covers: 16 MHz, because the SX1262 is the limiting device, and the W5500 (which
supports up to 80 MHz) and the SD card are configured down to match. A shared bus runs at the speed
of its slowest participant unless you’re willing to reconfigure the controller between transactions.
Follow-ups worth having ready: each peripheral needs its own chip select, so the GPIO budget
is part of the decision; SPI mode (CPOL/CPHA) must be compatible across devices sharing the bus; and
if one device genuinely needed to run faster, the answer is a second SPI peripheral, not
reconfiguring on every transaction. Also: SD cards can hold the bus during internal operations, which
is an argument for keeping them off a bus shared with a latency-sensitive radio.
Q8. [Intermediate] Why no external flash chip?
Strong answer covers: the ESP32-C6-WROOM-1-N8 has 8 MB on-module — enough for dual OTA
partitions plus a filesystem, which is the actual requirement (an OTA-capable field device needs room
for two images plus config/logs). A W25Q64JV was evaluated and rejected as unnecessary parts, board
area, and another SPI chip select.
The reasoning to state: don’t add a part until the memory map says you need it. Adding it “just
in case” costs BOM, area, a CS pin, and another thing to get wrong — and the module already answered
the question.
Q9. [Advanced] 🔥 The board has four connectivity options. How does it decide which to use?
Strong answer covers: that’s a firmware policy question, not a hardware one — the hardware’s
job is to make all four simultaneously available and independently power-manageable. The sensible
policy is a preference ladder by cost and power: Ethernet (free, powered, most reliable) → Wi-Fi →
cellular (metered, power-hungry) as the fallback, with failover driven by actual link health rather
than by carrier state, and hysteresis so it doesn’t flap between links.
The hardware consequence worth naming: because cellular is the fallback, it spends most of its
life idle — which is precisely why its power rail is separated (see
02-power-and-rf.md Q4): the rare, violent transmit spike must not disturb
whatever else is running.
Q10. [Intermediate] Draw me the block diagram.
Strong answer covers: power in (PoE / USB-C / barrel) → ORing auto-switch → 5 V rail → 3.3 V LDO
for the digital domain and a separate 3.8 V LDO for the modem. ESP32-C6 at the centre: SPI out to
SX1262, W5500 and microSD (each with its own CS); UART to SIM7080G; GPIO for the RF switch
control, status LED and user button; USB-C for power and USB-Serial/JTAG; SWD/JTAG header for
debug. RF: SX1262 → PE4259 SPDT → single LoRa antenna; W5500 → MagJack; SIM7080G → cellular antenna;
Wi-Fi/802.15.4 via the module’s own antenna.
The habit this demonstrates: being able to draw the interfaces from memory is what proves you
designed it rather than assembled a reference design.
Q11. [Beginner] What are the passives actually for?
Strong answer covers: three jobs, and being able to name them separately matters —
decoupling (a small ceramic at every IC supply pin to serve fast transient current locally, plus
bulk capacitance for slower demand), pull-ups/pull-downs (defining the state of resets, chip
selects, and mode pins so nothing floats at power-on), and termination/matching (series
resistors on fast lines, RF matching networks, and the RJ45 centre-tap network). The failure mode
worth naming: a floating reset or boot-mode pin gives a board that works on the bench and fails
intermittently in the field.
Q12. [Advanced] 🔥 If you had to cut cost by 30%, what comes off the board?
Strong answer covers: the answer should be feature-driven, not part-driven. Cellular is by far
the biggest single cost — modem, SIM slot, its own regulator, its own antenna — so the honest first
cut is “does this deployment need a cellular fallback?” Second, PoE: the PPS23730 plus magnetics is
expensive, and a passive-PoE or barrel-jack-only variant is much cheaper. Third, the module-versus-
bare-chip decision flips at volume, once certification is amortised.
The framing that lands: on a board like this the cost is in the radios and the power tree, not
in the passives — so cost reduction is a product-scope conversation before it’s an engineering one.
Software engineering
Infrastructure & security
Computer science
Hardware & engineering