Slide 1: Wireless Headset Technologies and Trends
Wireless Headset Technologies and Trends The proliferation of in-ear Active Noise Control September 17th, 2018
Slide 2: CEVA IP Portfolio
CEVA IP Portfolio AI Processors scaling from IoT to Automotive Handsets Base Stations Powered by Imaging IoT Computer vision Deep learning Sound DSPs Wi-Fi: 802.11n/ac/ax :AP & client Voice front-end SW Bluetooth 5: dual mode & BLE Speech recognition Audio playback Sound analytics Licensing & Royalt
Slide 3: CEVA's Unique Sound Processing Offering
CEVA’s Unique Sound Processing Offering DSP Software AI CEVA-TeakLite-4 for ClearVox package for NN Libs Ultra Low Power front-end voice pickup DSP NN Compute CEVA-X2 for high Noise reduction library Beam forming performance Acoustic echo cancellation Audio/Voice Codecs CEVA offers a comprehensive s
Slide 4: Headset Trends (earbuds, headphones, hearables …)
Headset Trends (earbuds, headphones, hearables …) Transition from 3.5mm jack creates opportunity for premium audio solutions for the powered (wired/wireless) headsets: ~1 Billion headsets in 2021 CEVA & Silentium Proprietary Information 4
Slide 5: CEVA Sound DSP Headset Solution
CEVA Sound DSP Headset Solution TeakLite-4 DSP Low-power always listening High performance voice pre-processing 1-3 mics noise reduction and beamforming Active Noise Control Voice trigger/commands/biometrics Low power audio playback / Audio over BLE Sound Neural Networks (audio analytics) Ultra-Low-
Slide 6: TeakLite-4 DSP Architecture
TeakLite-4 DSP Architecture Feature Configuration Pipeline stages 10 SIMD [bit] 64 DSP arithmetic data types [bit] 16/32/64/72 MAC [16x16-bit] 2 or 4 MAC [32x32-bit] 1 or 2 Single Precision Floating-Point Optional Data Memory width [bit] 64 or 128 Dedicated audio instructions Yes World’s Most Deploy
Slide 7: CEVA Sound Software Headset Solution
CEVA Sound Software Headset Solution ► ClearVox Voice Front-end Software 1. Noise Reduction 2-3 Microphones ► Multi channel Noise Reduction including SW VAD and Beamforming ► 2-3 MICs 2. Acoustic Echo Cancellation Featured-based Noise Acoustic Echo SW VAD Reduction Canceler ► Mono and Stereo Noise S
Slide 8: ClearVox Headset Noise Reduction
ClearVox Headset Noise Reduction ► Supported features ► Configurable 2-3 microphones ► Configurable microphone topology ► SW VAD (Voice Activity Detection) ► Fixed Direction of Arrival ► Dynamic Noise reduction ► Beamforming ► Different noise environments: white, directed, diffused, etc. CEVA & Sile
Slide 9: ClearVox Headset Acoustic Echo Cancellation
ClearVox Headset Acoustic Echo Cancellation ► Barge-in enabler ► Barge-in during music playback Acoustic echo ► Barge-in during personal assistant response ► Clean voice call enabler ► Intelligible voice in BT hands-free mode ► Supported Features: ► Configurable tail length ► Mono/stereo echo cancel
Slide 10: CEVA Sound NN Headset Solution
CEVA Sound NN Headset Solution ► “Sound NN at the edge” is a major trend ► In Headsets it is used for environmental sound sensing and speech recognition ► NN use-cases on CEVA TeakLite-4 sound DSP ► Voice trigger ► Voice commands ► Voice biometrics ► Sound sensing ► Customer proprietary NNs ► All NN
Slide 11: CEVA Sound NN Compute Library
CEVA Sound NN Compute Library CEVA Neural Network Library is a set of highly optimized neural network building blocks CEVA Neural Network Library is carefully composed to facilitate easy building and deployment of neural networks, with an emphasis on audio data Library functions are called “layers”
Slide 12: CEVA Bluetooth Headset Solution
CEVA Bluetooth Headset Solution Unique and comprehensive offering for both Bluetooth low energy Bluetooth dual mode (BLE) (BLE + BR/EDR) Bluetooth Low Energy (BLE) & Bluetooth dual mode (BTDM): Baseband controller Software protocol stack Modem & Radio Integrated platform with embedded RISC-V process
Slide 13: Audio over Bluetooth Dual Mode
Audio over Bluetooth Dual Mode BT Dual Mode remains the optimum choice for wireless audio Audio-over-BLE still an evolving technology CEVA’s RW-BTDM5 is the only viable Bluetooth 5 Dual Mode IP available 30+ Bluetooth Dual Mode design wins, many in mass production Additional challenges (& opportunit
Slide 14: One Stop Shop Ultra-Low-Power Sound Solution
One Stop Shop Ultra-Low-Power Sound Solution Bluetooth 5 Solutions Sound Solutions VAD Input Beamforming AEC Voice Trigger Output Ultra low power PDM to PCM Multi-mic Acoustic Echo Deep Neural Opus encoded Beamforming Canceller Network VT query Acoustic “What is the echo weather in San Diego?” OK Go
Slide 15: Silentium – Noise control across industries
Silentium – Noise control across industries Wearables Appliances Automotive HVAC Transportation Military Heavy Industry IT CEVA & Silentium Proprietary Information 15
Slide 16: ANC Signal Processing Controller
ANC Signal Processing Controller ASIC DSP IP CORE • Resource constrained • Scalable Performance performance • Optimal integration with Audio • Limited flexibility during • Optimal system level BOM the design process cost • Scalability via BOM cost • Few design decisions upfront • Limited system inte
Slide 17: Controller Architecture
Controller Architecture Hybrid FF/FB Feed Back 2 Microphones Feed Forward 1 Microphone 1 Microphone CEVA & Silentium Proprietary Information 17
Slide 18: ANC Behavior and Performance
ANC Behavior and Performance • Dynamic adaptability • Control Interfaces • Acoustic leakage from tips/foam • Dynamic noise sources, noise location, Noise PSD • Device user Interface Inputs • Static adaptability • Phone App • Acoustic human physiology variations • Sensors • Audio source/type variabil
Slide 19: Necessary Conditions for ANC
Necessary Conditions for ANC t(air) t(controller) < t(air) Power MIC ADC Filters/Algo DAC Driver Amplifier Error correction signal t(controller) t(controller) = Group Delay Acoustic path time ~ .050m/340m/s Reference = 147uS signal ANC Controller CEVA & Silentium Proprietary Information 19
Slide 20: Necessary Conditions for ANC
Necessary Conditions for ANC t(air) t(controller) < t(air) Power MIC PDM Filters/Algo DAC Driver Amplifier Replace ADC with a Digital t(controller) Microphone • t(controller) = Group delay • ANC performance only possible above 250Hz due to microphone latency ANC Controller CEVA & Silentium Proprieta
Slide 21: Challenges of Intra-Aural ANC
Challenges of Intra-Aural ANC Bone • Acoustic path time ~ .025m/340m/s = 74uS Conduction • Silentium predictive algorithms are required to Microphone compensate for short Acoustic Path time Packaging • More stringent ADC/DAC requirements • Silentium offers maximum broadband Short acoustic performanc
Slide 22: Milestones for an ANC ready SOC
Milestones for an ANC ready SOC SOC SOC SOC Product FPGA or initial Target Critical Production feature test samples Performance Release definition Parameters Microphone type- ADC Latency/SNR Initial Library Production intent analog/digital I2S Port Speed verification ANC Lib – V&V Mechanical form Cl
Slide 23: End Product – Development Milestones
End Product – Development Milestones Acoustic Product Digital Audio Release/ components features and Architecture Alpha build selection and EU definitions modeling Audio sources Driver & microphone Clocking, digital audio End-Of-Line for ANC. Characterization. design. calibration strategy. Power req
Slide 24: ANC Performance Continuum
ANC Performance Continuum • Predictive Filters - are employed to Tips mitigate longer group delays. Isolation /Comfort • “Soft” ANC - unmatched capability in design and flexibility Group Noise • MIPS vs. design goals adjustment. delay Attenuation • Flexible architectures - balance SOC and end produc
Slide 25: Calibration and Characterization
Calibration and Characterization Headphone PCBA • Acoustic path modeling. PC Digital Comm. • Simulation environment for ANC performance prediction. Tuning Tool ANC Library GUI • Performance Simulation and Coherence Evaluation. Control Signal Module Processing Module • Proprietary calibration strateg
Slide 26: Steps to a successful customer demonstrator
Steps to a successful customer demonstrator Identify demo goals and constraints: ANC, Audio, MIPS, power ANC library optimization with an FPGA version of a target SOC Identify necessary external components ADC, DAC, PA, etc. PCBA level integration Drivers, APIs, BSP for the full system with ADC, MIC
Slide 27: Silentium ANC Key benefits
Silentium ANC Key benefits Full OEM/ODM Flexible Low MIPS Algorithms and Memory product support Broadband Strong Partnership Performance with DSP core IP CEVA & Silentium Proprietary Information 27
Slide 28: CEVA & Silentium ANC Solution
CEVA & Silentium ANC Solution Based on CEVA ultra-low-power sound DSP Fully adaptive broadband ANC technology Modular ANC solution: Feedforward/Feedback/Hybrid Suitable for multi-noise environment Joint ANC demo is available Robust and fully integrated ANC solution for headset designs CEVA & Silenti
Slide 29: Q&A Wireless headset technologies and trends
Q&A Wireless headset technologies and trends The proliferation of in-ear Active Noise Control Youval Nachum: youval.nachum@ceva-dsp.com Senior Product Marketing Manager Igor Kofman: igor.k@silentium.com Director Product Management & Business Development, North America September 2018