Step-counter firmware for a BLE rehab wearable, with an accuracy study

Patients recovering from a knee replacement are asked to walk a little more each day. Physiotherapists want to see how much they actually walk, but consumer fitness trackers undercount badly at the slow, uneven pace of early recovery. In one audit at the Leeds trust, a popular wrist tracker missed about 40% of steps for patients using a walking frame.

Cadence Health's device clips to the waistband. It has an nRF52-class Bluetooth Low Energy microcontroller, a 3-axis accelerometer sampled at 25 Hz and a 100 mAh battery. The product lead's targets are: - step count within 5% of a manual count for normal walking at about 1.2 m/s; - within 10% for slow walking at 0.4 to 0.6 m/s, including with a walking frame or crutches; - no more than 20 false steps per hour while sitting, typing or riding in a car; - at least 7 days between charges, with the step count sent over BLE when the patient's phone app connects.

You do not need the device. You will record your own accelerometer data with a free phone app such as phyphox, clipping the phone to your waistband, and count steps by hand or from video. You will design the algorithm in Python, port it to fixed-point C that would run on the microcontroller, define the BLE service, and estimate battery life from datasheet values. An optional run on an ESP32 with an MPU6050 in Wokwi is welcome but not needed.

Done looks like a labelled dataset with a protocol others can repeat, an algorithm with accuracy numbers by walking speed, a C implementation that matches the Python version, a GATT specification and power budget, and a validation report the physiotherapy team can read. Be honest about where the algorithm still fails. This data will inform clinical decisions, so overclaiming is the worst mistake you can make.