Pico satellite (CANSAT)

A two-board pico satellite flown to the national finals of CANSAT INDIA. Sole electronics engineer, blank schematic to recovered hardware.

Flown CANSAT INDIA 2023 9 min read

Role

Only electronics hardware engineer

Ownership

Sole · schematic to flight

Scope

Power board · sensor and telemetry motherboard

Constraints

Fixed cylindrical envelope · mass and volume budget · ascent vibration · single deployment

Outcome

Launched and recovered · live telemetry throughout descent

Fabrication

JLCPCB · his own manufacturing outputs

His own work. Real geometry.

What it is

CANSAT INDIA is run by ISRO and the Astronautical Society of India. The team reached the national finals. Two custom boards sit in a stacked assembly inside a fixed cylindrical container: a power board, and a sensor and telemetry motherboard.

The power board takes a 2S Li-ion 21700 pack into two buck converters generating 5 V and 3.3 V rails, plus an ESC channel for the descent actuator and a servo channel, with a ribbon interconnect to the upper board.

The motherboard is a Teensy 4.1 host with a BNO055 IMU, a BMP390 barometric altimeter and a NEO-8M GNSS on I²C, an ESP32-CAM for imaging, and an XBee S2C Pro with an external antenna for the telemetry downlink. Coin cell backup, state LEDs, and a buzzer driven from a 555 astable for post-landing recovery.

Both boards use custom non-rectangular outlines to fit the container, on threaded nylon standoffs, inside a 3D-printed frame sealed with PTFE film.

Decisions

Kind

Design decision
Problem found
Verification

Problem found U1 · integration

Sensor crosstalk inside a sealed container

Found: During integration testing the IMU and barometer readings were corrupted once the boards were mounted inside the sealed container. Outside the enclosure they were clean.
Did: Treated it as a coupling problem rather than a firmware one — the enclosure was vibration-heavy and the sensor traces ran close to the switching and motor wiring. Added board-level shielding over the affected region and rerouted the sensitive traces away from the noisy nets.
Result: All sensors read within spec, and held through the flight.

Design decision U1 · layout

A four-layer board built, and then rejected

Found: Produced a four-layer version of the motherboard with a stitched via fence around the perimeter, intending to improve ground return and reduce the coupling.
Did: Did not use it.
Result: It did not fit the available area once the mechanical envelope was fixed, and it did not resolve the EMI behaviour of the sensor modules themselves, which were the actual source. The two-layer board with local shielding was the better answer for the deadline.

What I would do differently now

Written against my own board

The most useful thing on this site is the distance between this list and the work above it.

Power stage

Used an older non-synchronous buck regulator with a large through-hole toroidal inductor. I would use a modern synchronous converter now: better efficiency, much less heat in a sealed enclosure, a fraction of the board area. On a battery powered payload, efficiency is endurance.
The switching loops are larger than they need to be. I now place the input capacitor, high-side switch and inductor to keep the hot loop tight before anything else is routed.
There is no deliberate input filter. I would design one: common mode choke, differential mode capacitance sized against the switching frequency, and a TVS clamp chosen against the regulator's absolute maximum rating. Exactly the thing that would have reduced the coupling I later had to shield against.

Layout and stackup

Two layers with routed ground was the wrong starting point for a board carrying an IMU, a GNSS receiver and a radio next to switching converters. A four-layer stackup with a continuous ground plane should have been the default rather than the fallback.
I would keep the noisy power domain physically separated from the sensor domain, with a defined return path, instead of relying on shielding to fix it afterwards.
No controlled impedance was specified anywhere. For the radio and the GNSS antenna feed that was luck rather than design.

Process

No formal design review happened, because there was nobody to review it. I now run structured reviews and write down what was checked, what was found and what changed. The crosstalk problem would very likely have been caught on paper.
Component selection was not documented. I now record the trade-off and the lifecycle status for every significant part.
I did not simulate anything. I would now model the converter and the load transient in LTspice before committing to fabrication.

Verification

What I would do differently now

The system worked, the engineering judgement was thin, and I know specifically where. Most of what I would change, I learned by doing it properly at work over the last year.

On a battery powered payload, efficiency is endurance.

Other boards