NXP Adds to AI Automotive Portfolio With Audio Processing DSP Chips

NXP continues to advance automotive edge AI with the new SAF9000 and SAF91000 AI audio chips.

 

NXP envisioned an infotainment system based on the SAF9XXX family. 

 

The new family of chips contains an embedded MCU and multiple coprocessing cores for AI digital signal processing (DSP), machine learning (ML), and neural processing. The chips are connected and can include a five-tuner, software-defined radio (SDR) to deliver a connected AI audio system in midrange and high-end cars.

 

AI-Optimized Audio Digital Signal Processing

The SAF9XXX chips integrate AI and ML with an advanced AI-oriented DSP engine. The audio DSP uses a pair of Cadence Tensilica HiFi 5 DSP cores with large, fast-access onboard memory. These DSP cores contain optimized AI functionality and neural network algorithms. They work in concert with the MCU core and peripheral sections for security and safety, audio I/O, audio processing, and flexible connectivity. 

 

SAF9000 single-chip radio and audio processing solution.

 

The SAF9000 adds five software-defined tuners to the processing set and a radio processing section. The cores support all current broadcast standards such as AM, FM, DAB, DAB+, DMB, HD Radio, and CDR, including DRM for AM and FM (DRM+) bands. While software controls the number and configuration of tuners, all tuners can use the same hardware variant because of its wideband tuner concept.

Audio I/O includes integrated sample rate converters, 12 software-configurable audio analog-to-digital converters (ADC), and ASIL-A audio safety support. This delivers fully programmable digital audio input/output support for a large variety of digital audio interface formats.

 

SAF9100 scalable audio DSP with AI and ML capability. 

 

The SAF9100 includes the processing capabilities of the SAF9000 but foregoes the SDR tuners, so its advanced audio capabilities can integrate with other radio solutions.

 

Third-Party Compatibility

NXP designed its new SAF9XXX chip family to capitalize on third-party and open-source components—one of which is Audio Weaver by DSP Concepts, a development, debug, and deployment platform for programmable audio systems. Audio Weaver works with IP plugins from many audio library developers such as Bose, DTS, and Intelego. These plugins add AI and ML to enhance noise cancelation or noise enhancement for quiet electric cars. They can also alert telephony, music, communications, and immersive sound. By using the platform, the SAF9XXX family can leverage extensive development work by experts in a wide variety of automotive audio fields. The system includes field upgradability.

The SAF9XXX family uses an integrated Arm Cortex M7 for middleware, tuner and audio control, and other generalized processing needs. The chips have two general-purpose SPI ports and two PCIe ports. PCIe one is 3.1 compliant with data rate support of 2.5 Gb/s and 5.0 Gb/s. The second PCIe can be used as standard PCIe or Gbit Ethernet SGMII.

 

A New Generation of AI Infotainment

Infotainment has come a long way from the humble AM/FM car stereo of the twentieth century. A great system of that era came with multiple speakers, a compact disc player, and a high-power analog audio amp. Today’s systems include all of that, except perhaps the CD player, but also support fast system interconnection, demanding self-learning speech and voice functionality, and advanced adaptive sound control.

AI/ML in the chips listen to and interpret sounds in the vehicle's passenger compartment. Over time, the ML capability can distinguish between passengers and the driver by characterizing their voices, accents, and intonation. AI chips can also recognize environmental sounds, such as emergency vehicle sirens, while intelligently reducing unwanted sounds, like road and wind noise, in real time. All of this can be used to inform the driver and vehicle systems, leading to a safer and better driving experience.

 

---

 

All images used courtesy of NXP.