Designing a Car Computer: Sourcing the Hardware

The whole build starts with one stubborn constraint: the truck sits for days at a time, and when it sits, nothing is supposed to be awake. This is a computer for a 1969 Chevy Blazer (and a second truck, a Dodge). It sleeps until my phone comes near, wakes, boots, and talks. Offline-first, low power, always-on sentry that draws almost nothing and a real computer that stays dark until it is needed.

I am not going to walk through the software here. This post is about the part of the project nobody writes up: sourcing the hardware. How each part got researched, why one variant beat another, where it came from, and the traps that ate an afternoon. If you are planning anything similar, the traps are the useful part.

The vendor map

Almost every choice came down to “who actually has the real part, in stock, and will say what it is.” That sorted the build into four lanes:

• Adafruit (US) for the Raspberry Pi, the power HAT, the microcontroller, and the small breakouts. They stock genuine parts, publish real datasheets, and the product pages warn you about the gotchas.
• Data Alliance (US) for RF connectors and antennas. This turned out to be the most important sourcing decision of the whole build (see the connector trap below).
• Amazon for commodity: the DC-DC buck, tools, soldering supplies, SD cards.
• Local / automotive for the fuse, 12 V wire, the ignition-sense optocoupler, and the harness. Bin stock, bought by the unit, no shipping.

The rule I keep relearning: the fancy parts come from the specialist, and the boring parts come from wherever is closest. Trying to save two dollars on a connector by buying the wrong thing from Amazon cost more than the right part ever would have.

The power HAT, and a variant trap

The heart of the sleep/wake design is a power-management HAT on the Pi. I wanted the UUGear Witty Pi line: it takes 12 V straight from the vehicle, soft-switches the Pi, and shuts it down gracefully. Adafruit sells it in the US.

Two traps here.

First, the Witty Pi 4 is out of stock. That is the model the architecture doc originally named, and it simply was not available when I went to buy.

Second, and sneakier: there is a Witty Pi 4 “L3V7” variant, and it is not the same part. L3V7 is the 3.7 V Li-ion version. It takes 5 V in over USB-C – it is a battery UPS, not a wide-input vehicle supply. If I had grabbed it because the model number looked close, I would have bought a 5 V board for a 12 V truck.

The in-stock pick was the Witty Pi 5 HAT+: 6-30 V input (takes the vehicle 12 V directly, no separate buck in the Pi path), up to 5 A out, and it supports both the Pi 4 and the Pi 5. It is the vehicle variant the build actually needed.

The Pi 4 price shock

A note for anyone building in 2026: the Raspberry Pi 4 is not a $55 board anymore. The 4 GB model was about $120 at Adafruit when I ordered, and the 8 GB was around $190. DRAM pricing is the driver, and it is volatile – verify at the moment you order, because the number you read in a forum post from six months ago is fiction now.

I still bought the Pi 4, deliberately, over the Pi 5: it uses less power, and it keeps the 3.5 mm analog audio jack that the voice output path depends on.

The sentry brain

The always-on half of the system is a small microcontroller doing a duty-cycled BLE scan for my phone, watching ignition, and chirping. I picked the Adafruit ESP32-S3 Feather 8 MB with w.FL (product 5885): native USB-C, plenty of flash, and a connector for an external antenna – which matters because the entire truck body is steel and a PCB antenna inside a steel console is close to useless.

One small gotcha: the pin headers ship loose. You solder them on yourself. Fine with a Pinecil, but it is a “read the box” moment, not a “plug it in” moment.

The connector trap (the centerpiece)

If you take one thing from this post, take this.

The Feather’s tiny RF port is w.FL. That is the same thing as I-PEX MHF3. It is not u.FL. It is not MHF4. They look almost identical at a glance and they do not mate.

flowchart TD
    Port[Tiny RF port on the board] --> Q{Which series is it?}
    Q -->|w.FL| A[w.FL = I-PEX MHF3
CORRECT: W.FL-to-RP-SMA]
    Q -->|u.FL| B[u.FL
WRONG - does not mate]
    Q -->|MHF4 / IPEX4| C[MHF4
WRONG - cheap and everywhere,
but does not mate]
    A --> D[RP-SMA bulkhead + 2.4 GHz antenna]

Here is how the trap works. The Adafruit pinout page had stale text calling the port “u.FL,” which sent me looking for u.FL cables. Amazon listings are worse: plenty of them are labeled “u.FL” while showing something that looks like w.FL, and the cheap “MHF4 / IPEX4” antennas and pigtails – which are everywhere and cheap – do not mate with this board at all. You buy the wrong one, it arrives, and it physically will not connect. No adapter saves you; the series is wrong.

The fix came from going to a specialist instead of a marketplace. Data Alliance states it plainly, in the product copy, in the open:

“W.FL (HiRose) is equivalent to MHF3 (I-PEX).”

That one sentence is the whole resolution. The correct cable is a W.FL-to-RP-SMA. Not u.FL. Not MHF4. W.FL.

One more caveat that shaped the whole antenna plan: w.FL is rated for only about 20 mating cycles. This is a connect-once connector. I am not going to be plugging and unplugging this while I debug; I build the cable into the design and leave it alone.

The antenna assembly

The RF chain had to get the signal out through a steel case wall, so:

• A W.FL-to-RP-SMA-female cable from Data Alliance whose RP-SMA end is a bulkhead – it mounts through the case wall with a nut and washer, and the antenna screws on outside.
• A 2.4 GHz RP-SMA antenna on that bulkhead.
• A flat O-ring for the bulkhead (with the brass nut/washer it seals to IP68).

Two sourcing notes that matter. The cable is deliberately short – the thin 0.81 mm coax loses signal at 2.4 GHz, so every extra inch is range you throw away. And the bulkhead thread is 3/8 in, so the case wall has to stay thin (about 9.5 mm); buy the longer-thread version if your wall is thicker.

The DC-DC buck

The sentry needs its own 12 V-to-5 V converter. I went cheap and waterproof: a PlusRoc 12/24 V -> 5 V 5 A waterproof buck, sold as a 2-pack on Amazon, over a pricier Pololu module. The deciding feature was the USB-C output – it plugs straight into the Feather’s USB-C port, no terminal-block wiring in a cramped box. It is its own converter, deliberately: the other buck in the project is soldered to the dev Pi and is not free to steal.

The chirp

Parked feedback is a chirp, not speech – the amp and sub are ignition-gated, so a parked Pi has no audio path. That comes from a PAM8904 piezo driver amp and a large enclosed piezo, wired to the Feather with a STEMMA JST-PH cable. A bare piezo off a 3.3 V GPIO is weak; the driver makes it a real car-alarm chirp.

Small connector note: the Feather’s JST is STEMMA QT (JST-SH, I2C), which is a different connector from the 3-pin JST-PH the PAM8904 wants. Get the right cable.

What I considered and rejected

PiSugar 3 / 3 Plus, the bottom-mount pogo-pin UPS. It is genuinely clever – no GPIO header, no heatsink conflict, on-board RTC, graceful shutdown. Two reasons it lost: it is 5 V/3 A in and out only (a 12 V buck would still be required, where the Witty Pi takes 12 V direct), and it puts a Li-ion cell inside a hot cabin. The vehicle already has an auxiliary battery; that is the UPS. Adding a second battery in a truck that bakes in the sun is a safety problem looking for a place to happen. Rejected.

Takeaways

• US vs overseas sourcing is a real design input. Specialists for parts where correctness matters; commodity from whoever is fast and cheap.
• Out-of-stock and variant traps are the norm. The model number being close is exactly how you buy the wrong board.
• Connector naming is chaos. w.FL = MHF3, and no amount of squinting makes it true that it equals u.FL or MHF4. Trust the specialist’s plain-language copy.
• Prices are volatile. Verify at order time, and expect the Pi to have moved.
• Buy spares. I ordered three of the RF parts for two trucks; the parts are tiny, cheap, and a wrong or broken one otherwise stops the build for a week.
• Measure, do not guess. Every electrical value in this project is a guess until it is measured on the bench.

Where it stands

Everything is ordered or in hand. Both Pi 4 units are imaged and stress-tested (passed). Next: bench-verify the power path and the sentry wiring before anything goes live in the truck.

Parts

Approximate prices, USD. Some are from a multi-item order; “see order log” means the line item was not priced individually here.