Facebook Portal

Problem

Smart displays for video calling require exceptional audio quality in challenging acoustic environments: living rooms with TV noise, kitchens with appliance hum, open offices with background chatter.

Portal devices needed to:

• Capture clear voice while rejecting background noise (far-field microphone arrays)
• Deliver high-quality speaker output for video calls and media playback
• Maintain low latency for real-time communication
• Handle echo cancellation when speaker output feeds back into microphones

Approach

Led audio validation for Facebook Portal smart displays, focusing on:

• Microphone array performance: beamforming accuracy, noise rejection, voice pickup range
• Speaker system validation: frequency response, distortion, maximum SPL
• Echo cancellation testing: verify AEC (Acoustic Echo Cancellation) in realistic scenarios
• Integration testing: audio + video sync, firmware coordination

Test Strategy:

• Define audio requirements based on use cases (video calls, media playback, voice assistant)
• Build test methods for far-field voice capture (3m, 5m pickup distances)
• Create realistic acoustic test scenarios (living room, kitchen, office background noise)
• Coordinate with firmware team on audio processing pipeline (beamforming, AEC, noise suppression)

What I Built

Test Infrastructure:

• Far-field voice capture test rig (calibrated speakers at multiple distances/angles)
• Background noise playback system (real-world recordings: TV, appliances, office chatter)
• Speaker validation setup (anechoic chamber, APx analyzer, distortion measurements)
• Echo cancellation test system (loopback with controlled echo paths)

Documentation:

• Audio test plan with requirements traceability (PRD → test cases)
• Test case library organized by subsystem (mic array, speakers, AEC, integration)
• Firmware debugging guide for audio issues
• Manufacturing acceptance test procedures

Validation Process:

• DVT: confirm design meets spec across acoustic environments
• Build validation: verify consistency across production units
• Firmware regression testing: catch audio processing bugs
• Subjective listening tests: validate objective metrics correlate with perceived quality

Architecture

Diagram placeholder: Portal Device → Test Environment → APx + Measurement Mics → Python Analysis

Portal Device (DUT)
    ├─ Microphone Array (far-field voice capture)
    │   ↓
    │   Test Speakers (3m, 5m, with background noise)
    │   ↓
    │   APx Analyzer (capture quality metrics)
    │
    └─ Speaker Output (video call audio)
        ↓
        Measurement Microphones (APx)
        ↓
        Python Scripts (analyze FR, THD, SPL)

Outcomes

Qualitative Impact:

• Validated audio subsystem for on-time product launch
• Identified and resolved critical beamforming bug (off-axis voice rejection too aggressive)
• Built repeatable test methods that transferred to manufacturing
• Reduced customer audio complaints by ensuring consistent quality across units

What worked well:

• Realistic test scenarios (actual living room noise) caught issues lab tests missed
• Firmware collaboration accelerated debugging (test engineer + audio DSP engineer pairing)
• Manufacturing handoff went smoothly (clear acceptance criteria, documented procedures)

Challenges:

• Echo cancellation testing required custom scenarios (commercial tools too generic)
• Microphone array validation needed 3D positioning (anechoic chamber time-intensive)
• Subjective listening tests time-consuming but necessary (objective metrics don't tell full story)

Learnings

• Use realistic test scenarios early: lab-perfect conditions hide real-world issues
• Collaborate with firmware: audio bugs need fast iteration loops between test + DSP teams
• Document trade-offs: beam pattern tuning is always a compromise (directivity vs. sensitivity)
• Plan for manufacturing: DVT tests must scale to production (time, equipment, complexity)