Sleep tracking implementation in mobile app

Sleep Tracking Implementation via Mobile Application

Sleep tracking on phone without wearable — accelerometer under pillow or mattress. With wearable — heart rate + accelerometer for phase classification. Both work but fundamentally different accuracies and technical complexity.

Without Wearable: Accelerometer as Movement Sensor

Classic — phone placed on mattress. Body movements transmit through mattress to phone. Task: classify movement periods (restless sleep, awakenings) and rest (deep/light sleep).

Algorithm:

1. Record accelerometer at 10 Hz (battery-optimal, sufficient for detecting body movements)
2. Every 30 seconds — epoch: calculate activity_count = sum(|delta_x| + |delta_y| + |delta_z|) per epoch
3. Classification: activity_count < threshold → SLEEP, > threshold → WAKE
4. N-epoch patterns → determine sleep start and end

Activity threshold selected empirically — depends on phone placement, mattress firmness. First 3–7 nights — calibration via feedback ("I fell asleep at 23:15, correct?").

Sleep phase accuracy (REM/N1/N2/N3) — below 60% without heart rate. Movement only — rough division into "calm/restless". If product claims medical-grade — need wearable sensor.

With Wearable: HealthKit Sleep / Health Connect Sleep

iOS + Apple Watch:

Apple Watch automatically records sleep via own algorithms (accelerometer + heart rate + skin temperature on Series 8+). Result — in HealthKit as HKCategoryValueSleepAnalysis:

let sleepType = HKObjectType.categoryType(forIdentifier: .sleepAnalysis)!

let query = HKSampleQuery(
    sampleType: sleepType,
    predicate: HKQuery.predicateForSamples(withStart: startDate, end: endDate),
    limit: HKObjectQueryNoLimit,
    sortDescriptors: [NSSortDescriptor(key: HKSampleSortIdentifierStartDate, ascending: true)]
) { _, samples, _ in
    guard let samples = samples as? [HKCategorySample] else { return }
    samples.forEach { sample in
        switch sample.value {
        case HKCategoryValueSleepAnalysis.asleepCore.rawValue:
            // N1/N2 — light sleep (watchOS 9+)
        case HKCategoryValueSleepAnalysis.asleepDeep.rawValue:
            // N3 — deep sleep
        case HKCategoryValueSleepAnalysis.asleepREM.rawValue:
            // REM
        case HKCategoryValueSleepAnalysis.awake.rawValue:
            // Awake
        default: break
        }
    }
}

Before watchOS 9 only inBed and asleep — no phases. If app must show sleep phases from Apple Watch — minimum iOS 16 / watchOS 9.

Android + Health Connect:

val response = healthConnectClient.readRecords(
    ReadRecordsRequest(
        recordType = SleepSessionRecord::class,
        timeRangeFilter = TimeRangeFilter.between(startTime, endTime)
    )
)

response.records.forEach { session ->
    val duration = Duration.between(session.startTime, session.endTime)
    session.stages.forEach { stage ->
        when (stage.stage) {
            SleepSessionRecord.STAGE_TYPE_DEEP -> { /* deep */ }
            SleepSessionRecord.STAGE_TYPE_REM -> { /* REM */ }
            SleepSessionRecord.STAGE_TYPE_LIGHT -> { /* light */ }
            SleepSessionRecord.STAGE_TYPE_AWAKE -> { /* awake */ }
        }
    }
}

Recording Sleep (Own Data)

If app itself analyzes sleep (via phone accelerometer):

iOS — write to HealthKit:

let sleepSample = HKCategorySample(
    type: HKObjectType.categoryType(forIdentifier: .sleepAnalysis)!,
    value: HKCategoryValueSleepAnalysis.asleepCore.rawValue,
    start: phaseStart,
    end: phaseEnd
)
healthStore.save(sleepSample) { success, error in }

Smart Alarm

Most tracker feature — wake in "light" phase within window (e.g., ±30 min from desired time). Implementation on phone: app active in background, analyzes accelerometer, detects activity in alarm window — triggers via UNUserNotificationCenter (iOS) or AlarmManager.setAlarmClock() (Android).

AlarmManager.setAlarmClock() — only Android alarm guaranteed to fire on strict battery saver (Doze mode). Regular setExact() and setAlarmClock() may delay several minutes.

Typical Errors

• Not accounting for sleep breaks (person woke 20 min) — algorithm splits one session into two
• Forgetting daylight saving time — timestamps UTC, convert to local only for display
• Not filtering "sleep" < 30 min duration — naps or classification errors get into statistics

Timeframes

Basic tracker with accelerometer and alarm — 3–5 weeks. With HealthKit/Health Connect integration, sleep phases and period analytics — 6–10 weeks.