The Foundation for Taking IoT Devices Everywhere

One approach to satisfying demands for high-efficiency DSP performance and real-time function support

The key to success for IoT devices and systems is the ability to deploy devices wherever they are needed and be able to connect to the core network immediately with minimal intervention. There are countless applications in transportation, industrial control, agriculture and for consumers where the ability to communicate with the cloud to make sense of local sensor readings is vital. One wireless technology is key to delivering on that aim: cellular.

Figure 1: Cellular IoT for the Massive Internet of Things

Figure 1: Cellular IoT for the Massive Internet of Things

Examples of systems that need the flexibility of connectivity that cellular offers include asset trackers. Public transport vehicles already use GPS-enabled tracking systems to report their position to a central controller—allowing operators to keep travellers updated on when their next ride will arrive. But the availability of flexible, low-cost connectivity can extend location tracking to many more pieces of equipment. A common source of delay in construction projects, for example, lies in equipment not being in the right place. A sensor node on the machinery that has access to cellular connectivity can report its position every time it is moved without the owner having to deploy their own private wireless network or worry that the equipment will move out of range.

Constant Connectivity
Wireless operators and the 3GPP standardisation group have embraced the vision of constant connectivity. They have accelerated plans to not only bring IoT capability to the forthcoming 5G networks, but also to provide a stepping stone to universal low-power wireless with the new Cat-M1 (eMTC) and Cat-NB1 (NB-IoT) LTE categories ratified by 3GPP last year.

The IoT-focused cellular standards are intended to support the needs of battery-powered sensors and devices that send small amounts of data intermittently. The protocols make it possible to maintain a long-term connection without forcing the device to respond to the basestation at regular intervals. Changes such as this keep energy usage to a minimum. But device integrators will only reap the full benefits of these new protocols if they are able to take advantage of low-energy processing throughout the design.

There are four requirements for IoT devices that are becoming crucial. Sensing and connectivity are already established as key to the core functionality of IoT devices. The ability to sense position using GPS, often in combination with triangulation based on wireless communications, is important for devices that are installed in vehicles and even for nodes intended to reside in a fixed location.

In agricultural applications, for example, multiple sensors may be scattered across fields to monitor environmental conditions, supporting the efficient use of irrigation in drought-prone areas. By making the sensors position aware through GPS reception, they do not need to have their location encoded manually. And the device can report when it has been moved either accidentally or maliciously.

Lacking the surface area to present more than a minimal touch interface, a user interface based on voice provides the ability to interrogate and control an IoT device locally. Home-based systems have demonstrated the efficacy of local voice detection and preprocessing algorithms coupled with cloud-based speech recognition, helping to offload the most compute-intensive functions from the IoT device. A voice-detection feature may be just as important in the industrial setting, letting sensors respond to voice commands when a maintenance technician is nearby even when they are hidden away behind other machinery.

With the inclusion of a voice interface and built-in cellular communications, smartwatches and similar designs can run and support user commands without needing to be linked to a smartphone. This makes smartwatches much more viable for healthcare and fitness applications by allowing the user full access to the wearable’s functions without needing to carry a larger electronic device.

Hybrid Functionality
The combination of functions leads to a demand for high digital signal-processing (DSP) performance at high efficiency, combined with the ability to run control-intensive code to support real-time functions. A traditional approach to this combination of functions would be to use dedicated processors for each function. But this increases silicon cost and can increase power consumption because of the need to run two or more processor cores at once passing data between them. The need to synchronize threads can often lead to cores idling while the other finishes its tasks, which increases the amount of energy lost through leakage. A single-core solution that balances control and signal-processing workloads provides opportunities for fine-grained power management and better efficiency.

In the design of the CEVA-X1 processor core, the architects paid close attention to the needs of IoT devices. Their solution was to design the processor architecture and C compiler together to achieve superior C code generation performance and compactness with a very long instruction word (VLIW) single instruction multiple data (SIMD) architecture with variable length instruction encoding. CEVA-X1 employs a four-way VLIW architecture but couples this with an instruction pipeline that is highly responsive to the branch-intensive code found in real-time control algorithms and in cellular protocols. When running the EEMBC CoreMark benchmark, which targets control-oriented applications, the CEVA-X1 achieves a score of 3.6 CoreMark/MHz—providing comparable efficiency to processor cores designed purely for this purpose. But with a multiply-accumulate engine able to run two 16×16bit calculations in parallel and the ability to run up to four instructions in parallel as a VLIW machine, the CEVA-X1 provides the instruction set richness and pipeline throughput required for advanced signal-processing. To help reduce code size, commonly used instructions can use a compact 16bit encoding. Byte-level addressing allows the space-efficient ordering of data in memory.

Figure 2: CEVA-X1 processor architecture

Figure 2: CEVA-X1 processor architecture

The CEVA-X1 supports flexible processing through its implementation of a variable-length pipeline with up to ten stages. The full ten stages are deployed when the processor is running more complex DSP instructions. This provides a pipeline that is more responsive to code that intersperses DSP-intensive sections with branch-heavy execution.

To reduce the penalty of branching code, the CEVA-X1 employs dynamic prediction as well as providing support for predicated execution in many instructions so that simple if-then conditions can be executed inline and avoid the possibility of interrupting instruction flow.

Context Switching
Context-switch performance in a device that needs to juggle communications with sensor processing and voice detection is as vital to overall responsiveness as branching support. The CEVA-X1 supports low-latency context switching through the use of a fast register switching and the inclusion of multiple stack pointers. This ability to move between contexts using on-chip resources avoids the need in many cases to force register contents into main memory. A complete thread context can be stored and recalled almost immediately, allowing the ability to switch rapidly between tasks and respond to changing conditions in the system.

Figure 3: Asset tracker architecture.

Figure 3: Asset tracker architecture.

Although the CEVA-X1 has the ability to support the Cat-NB1 protocol entirely in software and execute sensor-processing functions, standard specific instruction extensions for Turbo and Viterbi processing in addition to encryption and correlation are available to optimize performance and reduce power consumption. A complete Cat-NB1 modem including protocol stack and PHY processing can run concurrently with GPS and sensor-fusion code on a CEVA-X1 at a clock speed of just 150MHz.

Through its efficient support for connectivity, control and DSP, the CEVA-X1 provides the ideal foundation for the rapidly emerging market for cellular-enabled IoT devices, taking full advantage of the ability to connect to the Cloud from almost anywhere.

emmanuel_gressetEmmanuel Gresset is a Business Development Director in CEVA’s Wireless Business Unit. For the last 30 years, Mr. Gresset has been with systems and semiconductor companies working in the fields of signal processing, wireless modems as well as processor and system-on-a-chip architecture in various companies: Octasic, STMicroelectronics, Philips, VLSI Technology, Spectral Innovations and Thomson. He received his M.Eng from the Ecole Supérieure d’Electricité in Paris.

Share and Enjoy:
  • Digg
  • Sphinn
  • Facebook
  • Mixx
  • Google
  • TwitThis