June 05, 2014Pat Lawlor
Sensor processing requirements are only increasing. The simple sensor processing days of the past are gone forever as we now see the trends of more sensors and more algorithms to enable breakthrough experiences. The logical question that arises is: what is the most efficient way to deliver these sensor-based experiences in power constrained mobile devices?
The short answer: smart integration of a specialized sensor engine by taking a heterogeneous computing approach. In a recent webinar, Qualcomm Technologies took a deep dive into how to design and enable the optimal sensor solution. Here are my key two takeaways:
A digital signal processor (DSP) is the right specialized engine architecture for sensor processing. As discussed in my previous blogs, the fundamental premise of heterogeneous computing is that running the appropriate task on the most suitable specialized engine delivers performance and power benefits. As it turns out, processing sensor data has diverse requirements such as high performance, real-time processing, and always-on sensing—and all of this at low power.
The high performance is not just any type of performance. By examining a typical sensor algorithm, such as course motion classification, it becomes clear that the processing tasks are a combination of signal processing, control flow processing, and real-time processing. The right architecture (or tool) for these processing tasks is a DSP, such as Qualcomm Technologies’ Hexagon DSP, as opposed to a microcontroller (MCU).
The Hexagon DSP was designed to maximize work per cycle, which results in optimal performance per milliwatt and a wide range of performance efficiency. As an example of the Hexagon DSP’s efficient performance, the chart below shows how it excels at signal processing and control processing as compared with certain third-party MCU-based solutions.
Smartly integrating the specialized sensor engine into the SoC is the right approach for optimal processing efficiency. Because Qualcomm Technologies takes a holistic approach to system design, we are able to develop a smartly integrated SoC architecture with system-level innovations. Compared to using a discrete sensor hub, smart integration has two huge benefits: inter-processing engine communication and access to shared DRAM.
Inter-processing engine communication provides a direct path between the sensor engine and other processing engines in the integrated SoC. Unlike a discrete sensor hub, an integrated sensor engine can take advantage of inter-processing engine communication. This direct communication allows the integrated solution to bypass the CPU and let it stay asleep, which results in saving power, a faster response, and a better user experience.
Access to shared DRAM means that the integrated sensor engine has direct access to the DRAM, which acts as scalable memory. This is very useful since the need to support more sensors, more algorithms, and more use cases is on the horizon and will require additional memory. An integrated sensor engine can deliver a broad range of additional experience at essentially no additional cost. On the other hand, the discrete solution has no direct access to the DRAM and as a result is primarily limited to the RAM in the package. To deliver additional sensor experiences, you would need to buy a discrete sensor hub with additional RAM, which increases cost.
Overall, smart integration provides many benefits over a discrete solution such as reduced cost, development time, and power consumption, as well as more responsive and comprehensive user experiences.
Now you know why Qualcomm Technologies has smartly integrated a specialized sensor engine. It’s because that approach provides optimal processing efficiency. Be sure to check out the webinar for many more details.
Want to learn more? Look for future blogs and sign up for our newsletter to receive the latest information about mobile computing.
Heterogeneous Computing4Technical Marketingnonenone
0June 05, 2014