Vector DSP IP charts course for IoT/M2M

For some time, we’ve been talking about ideas for IoT-specific chips, evolved from garden-variety MCUs or mobile SoCs. I sat in on a fascinating talk from an MCU vendor at ARM TechCon 2015 regarding multi-protocol radio silicon, and a question kept coming from the audience: what about software-defined modems? The vague response from the presenter was software-defined modems on an IoT/M2M scale is possible, but expensive in terms of silicon and power consumption.

That’s not a new problem. One of the motivating factors for adding fast multiplier capability to the ARM7 core way back when was to enable some lighter algorithms to run without a full-blown DSP. There are still jobs that need a high performance DSP, like 4G LTE, and the drive from CEVA, Qualcomm, and others has been to streamline DSP cores for efficiency.

A lot of territory exists between those extremes, much of it on the IoT and in M2M applications. LTE has been a two-edged sword for M2M. GSM bands are being repurposed for use in LTE as crowded spectrum becomes scarce. AT&T, with a burgeoning 4G base to worry about, is sunsetting its 2G network at the end of 2016. LTE has covered the globe, literally, at the expense of some other once-promising standards such as WiMAX. For those that need increased mobile bandwidth and a long future ahead, LTE is hard to beat in M2M.

Let’s suppose your ideal application doesn’t resemble a head-shrunken smartphone with a high performance LTE DSP core. There are a lot of IoT/M2M protocols to choose from. Some are quite solid, like Bluetooth, GNSS, Wi-Fi, and ZigBee. Some are gaining attention rapidly, most notably LoRa and Thread. Some are still dripping wet specifications, like Narrowband-IoT. Some are dedicated to IoT networks, like SIGFOX. The use cases vary widely.

Several questions arise. Who knows, for sure, which of these specifications will win, and in what exact variant of the specification? Maybe more importantly, will some edge devices be required to operate on more than one network, either simultaneously or with one side always-on and the other sleepy? What about chips for multi-protocol gateways? If one commits to a network, or combination of networks, and something changes, a hardware-centric design can blow up.

CEVA is challenging the idea that vector DSP can’t scale to these lower power IoT/M2M devices. Two new cores, the CEVA-XC5 and CEVA-XC8, go after these software-defined modems in style. Compared to the previous XC323 low-end, the new IP brings up to 70% lower dynamic power consumption, 40% die area reduction, and 20% lower memory usage.

There are certainly use cases where a tuned hardware implementation of a mature standard can be better than software defined modems. Bluetooth Smart comes to mind, and CEVA’s Emmanuel Gresset admits that if BLE is all a design needs, CEVA’s hardware-based implementations are better. Software and vector DSP is the play where there are more advanced IoT/M2M protocols and some flexibility required, maybe in regional deployments or where specifications wax and wane in popularity over a long life cycle. There are some details in antenna design that need to be addressed when combining radios, but those also exist in hardware-based designs.

I go back to the unfortunate story of WiMAX, a specification that was supposed to dominate wireless deployments and instead lost silicon supporters rapidly to LTE. While backers of all these new IoT/M2M specs are optimistic, history may repeat itself for things like Weightless and 802.22. The XC5 and XC8 support LTE Cat-0 and Cat-M, including long DRX and power saving mode (PSM). On the other hand, SIGFOX or LoRa or Narrowband-IoT may take off and be the next big thing. The name of this game is flexibility.

The primary difference between the XC5 and XC8 is the MAC capability – 16 per cycle in the XC5, 32 per cycle in the XC8, presumably with some area and power delta. Both feature 8-way unrestricted VLIW, and 4-way set associative non-blocking program cache. AXI integration is a given. The power scaling unit allows multiple clock sources and multiple voltage domains, allowing flexible SoC power management strategies.

This kind of power management capability also pushes this DSP IP down into wearable space, where some designers have chosen tethered operation to save power instead of providing a direct LTE connection on a wearalone. It also opens the possibility of multi-protocol software-defined modems, as in this example with LTE Cat-0 and GNSS. LTE-OTDOA is also gaining momentum, using basestation signals for positioning.

One reason software defined modems haven’t been in this IoT/M2M conversation much is until now, power efficient vector DSP IP hasn’t existed. The fact that the microcontroller vendors are looking long and hard at multi-protocol radios for IoT devices is significant – and they may get it right on some point implementations. For many others working with specifications that are in flux or of varying regional importance, a soft vector DSP-based implementation may be the best way to get there quickly and cost effectively.

For more on the CEVA-XC5 and CEVA-XC8 DSP cores, visit:

CEVA-XC5 / CEVA-XC8 Communication DSPs

More articles from Don…