Sensing, Processing and Connecting: IoT Fundamentals

Internet of Things, or Internet of Everything, is certainly the buzzword of the year. Does IoT describe one product family? Not really as the acronym describes a family of concepts, each of these concepts could effectively be turned into a family of products, if this concept reach the market, or fulfill a market need. Nevertheless we could unify IoT definition, saying that IoT fundamentals at hardware level are: sensing, processing and connecting. These three actions will serve as an IoT product foundation, the finished product also including software encapsulation and security. An IP vendor is not selling a finished system, but some of the hardware “lego” pieces that you need to integrate to build the system. CEVA is typically one of these IP vendors you may want to talk with when defining an IoT product, as the company propose two of these fundamental pieces: Processing (thanks to the DSP IP core family) and Connecting (thanks to Riviera Waves acquisition). If you consider that CEVA has developed their DSP IP port-folio by supporting the mobile phone semiconductor industry, you realize that the company should be well positioned to address the various needs of IoT systems by offering low power solution. Low power is the mobile industry DNA, it’s also CEVA DNA because CEVA’s IP port-folio development is tightly coupled with battery powered, mobile systems.

If you search for low cost, low power sensors, you will certainly find several products filling your requirements. However, low cost/power generally means noisier sensors. That’s why DSP fit in sensing, as you need a lot of signal-cleaning (filtering, smoothing, calibrating) in order to extract meaningful data. Moreover, with the introduction of mics and image sensors, no processor does a better job than a digital signal processor: DSP have been invented to process algorithms, and voice or image processing rely on algorithms.

Thus, DSP-based sensing is not only the best (algorithm) processing option, it’s also the best approach for power saving, as we can see on the above example of MP3 decode comparison. Main CPU is clearly the worst option (it’s normal as a CPU is supposed to be general purpose processor) and the power consumption of M4 MCU/DSP is still 2x this of an optimized DSP core like CEVA-TL4.

The next question is to know if you can support multiple sensing technologies by using a single DSP (or do you have to use a specific DSP for each sensing technology). In fact the answer is in the picture below: the same DSP core can support sensor fusion (contextual awareness + motion gesture + indoor navigation), voice trigger, ultrasonic gesture, face trigger and BlueTooth Low Energy beacons. CEVA is claiming that such a DSP solution implemented on 28HPM supporting Always-on (sensor fusion + Voice Trigger + Race Trigger + BLE) consume less than 150 uW…

CEVA propose using the CEVA-TL421 as an audio analyzer, supporting mobile wearable, smart home and robot applications, all these applications requiring supporting:

• Voice recognition and speaker identification
• Speaker separation through beamforming
• Environment sensing like in the train or cinema
• Emotion detection

To support video analytics, the CEVA-MM3101 Computer Vision Engine would be suited for mobile, wearable, smart home, smart cities and security & surveillance markets. For such markets the DSP need to support:

• Human & object recognition
• Face recognition, gesture recognition
• Tracking based on feature and pattern matching
• Emotion detection
• Image and video enhancement or Video stabilizing.

If you goal is to develop an IoT platform, to serve as a demonstrator as well as to support software development, you can put all together the various CEVA DSP core (CEVA-TL410, CEVA-TL421 and CEVA-MM3101) with the CEVA Connectivity Platform supporting Zigbee, Bluetooth and WiFi. But you will probably need to integrate a low-power CPU/MCU (don’t need a powerful one, as the DSP cores are running the high performance processing), shared memory and various interfaces like I2S, I2C, LCD drivers or some of the MIPI specification like CSI or SoundWire.

Eric Esteve from IPNEST –