Getting it done
To get the design done, the team ended up leaning on Nordic, the supplier of the nRF24AP1
single-chip 2.4-GHz transceiver with embedded ANT protocol. According to Mike Paradis, sales manager at ANT Wireless, leaning on Nordic was one of the best decisions Park and his team could have made. “The initial [RF]design was very rudimentary: Nordic was very helpful once they got involved,” he said.
Nordic’s radio is proprietary, versus standards based, hence it is completely optimized for low power, according to David Day, sales manager at Nordic Semiconductor, ASA.
The importance of the component vendor for any design cannot be overstated, whether it be for the RF, main processor or other components. “That was one of the big decisions we made as a small company,” said Park. “Some vendors are great, and it’s not a function of size, as small or large can be responsive,” he said. “Others can be complete black holes.”
According to Paradis, the best thing a designer can do is take the help offered by the vendor. “Go to the people who are the leaders: they have all the use cases,” he said. “For example, Nordic has a complete reference design for a keyboard and mouse – just make your own form factor.” As a result, Nordic has design wins with Logitech (mice), Nike, and Polar. For its part, ANT has the complete low-power protocol and with the ANT Files System it has critical issues such as device association -- without a user interface – already figured out, he added.
When it came time to select the main processor upon which the algorithm would run, “a lot of it was educated guesses,” Park conceded. “We had only a vague notion of the power requirements of the software algorithm.” By this time, Park had divided the team up into separate groups: RF, firmware and I/O display, and algorithm development.
The team did know they had to minimize processing power consumption and so they narrowed the choice down to MCUs at that time (2007) from Texas Instruments and Atmel. As is so often the case, the final choice came down to a device they were familiar with, in this case a TI 16-MHz MSP430 MCU
with 92 Kbytes of flash, 4 Kbytes of RAM and a 12-bit analog-to-digital converter. It also happened to have the exact amount of flash and RAM the system needed, said Park.
The recently published 2010 EETimes Embedded Market Study
showed that 48 percent of the respondents stick with a chip they’re familiar with, though the ecosystem and tools are the prime reason behind that original choice. In this year’s study, TI was by far the favorite on that basis alone, at 17 percent. Freescale and Microchip came in second and third, at 10 and 9 percent, respectively.
Fig.2: Inside the FitBit, showing RF, MCU and MEMS choices.
3D MEMS: digital not necessarily better
The third and final piece of the component puzzle was the choice of 3-axis MEMS accelerometer. While designers have a tendency to go digital instead of analog, Park suggests that designers take a closer look at analog devices as they’re sufficient for most applications, without the added cost and complexity of a digital option.
For the FitBit, power consumption was a priority, along with noise levels, as the algorithm is particularly sensitive to noise, said Park. As a result, the team opted for the older-model MMA7341L
3-axis accelerometer from Freescale.
“We had worked with an ADI [Analog Devices Inc.] part, but chose this as it was cheaper, smaller and had lower noise,” said Park, adding that it’s half the size now than it was at the time.