The body speaks in waves HealthTech · Signal Processing · AI/ML
BCG Channel 1 · PVDF · 250 Hz

The body speaks in waves.

Deep dives into the engineering behind wearable health tech — BCG signal processing, sleep-stage AI, sensor design, and the systems that make non-invasive monitoring possible.

Read Latest Browse Topics

What your body is telling you

19.8M

Heart disease is the #1 cause of death globally — every single year.

19.8 million people die from cardiovascular conditions annually. Most of these conditions are detectable early — the challenge is that the warning signals appear long before any symptoms do.

WHO / Global Burden of Disease 2022
10+ years

Heart disease builds quietly for over a decade before you feel anything.

By the time most people receive a diagnosis, the condition has been developing silently for years. Early detection — not just treatment — is what changes outcomes.

American Heart Association · ESC Guidelines 2023
1 Billion

Around 1 billion people have sleep apnea. Most don't know it.

Sleep apnea causes brief pauses in breathing during sleep, repeatedly straining your heart through the night. A simple sensor under your mattress can detect it passively — no clinic visit needed.

Lancet Respiratory Medicine 2019
Days early

Your heartbeat rhythm changes before you feel sick.

Heart rate variability (HRV) — the slight variation between each heartbeat — reflects how well your body is coping. It drops days before illness, burnout, or a cardiac event, giving you an early window to act.

Frontiers in Neuroscience 2020
< 5%

Almost nobody monitors their sleep health — even once.

Yet sleep is when your body repairs itself and your heart works hardest. A single night of data can reveal patterns that a routine checkup would completely miss.

Sleep Advisor Global Survey 2024
~$0

Passive nightly monitoring now costs nothing extra.

A hospital sleep study costs $3,000–$5,000 and happens once. A contactless sensor under your mattress tracks your heart rate, breathing, and sleep stages every night — automatically, with no wires or effort.

AntarSpand · Clinical cost comparison 2024

Latest writing

BCG · Signal Processing

Why your BCG J-peak keeps drifting: a field guide to R→J delay variation

After aligning hundreds of nights of ECG and BCG data, the ±50ms beat-level variation in R-to-J delay isn't noise — it's physiology.

Chinmay Bhagat 12 min read
AI / ML

Knowledge distillation: teaching a BCG model with ECG labels

When your teacher model is 85% accurate and your student input is noisier, you use that gap intentionally.

8 min read
Sensor Hardware

PVDF vs capacitive: which sensor survives a restless sleeper?

Two sensor modalities, one hostile environment. After six months of field testing, here is what actually breaks.

5 min read
HealthTech

The engineering stack behind contactless vitals: from PVDF film to sleep stage

Sensor physics, ADC design, embedded firmware, cloud DSP, and the ML inference layer that ties it together.

14 min read
HealthTech

BLE for overnight ECG: what the spec sheets don't tell you

Bluetooth Low Energy looks ideal for overnight biosignal recording. The spec sheet omits the parts that will break your study.

7 min read
Deep Learning

Transformers for biosignals: what the hype gets right (and wrong)

Attention mechanisms are genuinely useful for long-context physiological signals. The implementation details are where most papers quietly fail.

11 min read
BCG / HRV

Building a Signal Quality Index from scratch

An SQI is the immune system of your BCG pipeline. Here is how to build one that actually rejects bad data without being paranoid.

9 min read
BCG / HRV

Snoring detection from BCG: separating breath from ballistic noise

Snoring produces a characteristic vibration signature in BCG. Isolating it from cardiac and respiratory signals requires understanding all three simultaneously.

6 min read

The person behind the signal

  1. ECG–BCG Synchronization Pipeline

    Full-stack system aligning Polar H10 ECG with POD4 PVDF BCG. Cross-correlation global sync + R-to-J peak matching, ±50ms delay handling, drift logging, interactive alignment UI.

    • Python
    • NumPy/SciPy
    • asyncio/BLE
    • Polar H10
    • PVDF

  2. Sleep Staging Transformer (BCG)

    Transformer trained on SHHS for 4-class sleep staging per 30s epoch. Knowledge distillation from ECG teacher (85% acc) to BCG student model.

    • PyTorch
    • SHHS
    • Transformer
    • Distillation
    • HRV

  3. iOS Background ECG Recorder

    Swift app for overnight BLE ECG from Polar H10. Solved 7-year epoch offset bug, timestamp reconstruction, 5-min rotation, iOS background execution constraints.

    • Swift
    • CoreBluetooth
    • Background Tasks

  4. Signal Quality Index (SQI)

    Pre-training classifier for BCG segment quality. SNR + template correlation + Isolation Forest motion detection. Adaptive threshold replacing GMM.

    • scikit-learn
    • Isolation Forest
    • CWT/Ricker
    • 8–20Hz BPF