Hello again,
Sorry for the long delay since the last post, I hope that you find this post interesting enough to make up for the wait.
One of the most anticipated features of Bluetooth low energy will be the ability to run on lithium coin cells. Although some companies have launched regular Bluetooth products that do use coin-cells, most current Bluetooth products run on rechargable batteries (either Li-ion or NiMH) as the battery life usually is not long enough to jusify the use of primary (non-rechargable) batteries.
Conventional wisdom has it that coin cells have a hard limit on the maximum peak current that can be sourced, and this is usually given as 15 mA or 20 mA. However, peak duration is usually not mentioned, so it is hard to figure out how "peak current" actually differs from the maximum constant current load. Many battery datasheets are not much help, as they often provide data only for the constant current case. For Bluetooth low energy, the peak current draw is very short, usually significantly less than 1 ms. One of the few sources of information I have been able to find that provides good documentation about pulsed use provides much bigger numbers for maximum peak current (see the bottom of the page at http://www.renata.com/content/3vlithium/tech_engineering.php), which indicates that even 50 mA over 100 ms may be feasible.
To get to the bottom of this, our team put together a measurement setup to simulate a BLE pulsed load and measure the actual capacity we could get out of a CR2032 cell with this type of load. We ran the tests with peak currents of both 15 mA and 30 mA to see if there was any difference. The results were quite interesting.
First, I would like to touch on how a coin cell behaves. A coin cell may be modeled quite well by assuming it to have a relatively large internal resistance that increases as the battery is drained. There is also some internal capacitance, but that has limited influence. During pulsed operation such as in a typical BLE system, the voltage will drop during the current peak and return to the open-circuit voltage during rest periods, when the BLE IC only draws current in the uA range.
The figure above shows an actual oscilloscope plot of the CC2540 implementing a BLE slave. The CC2540 wakes up from sleep, does some pre-processing, receives a packet from the master, transmits an acknowledgement and goes down to sleep. The plot is heavilly zoomed in; the whole active phase is over in a couple of ms, while the time between events will typically be hundreds of ms. How does this type of current profile affect a coin cell?
The plot above shows the voltage measured across a CR2032 battery over the lifetime of the battery. The voltage under load as well as the open-circuit voltage drops over time as the battery drains. The plot is for the 30 mA load. The load profile used was 8 mA for 2 ms, 15/30 mA for 1 ms, then 8 mA for 2 ms. This is considered to be a worst-case BLE scenario, in most cases the times will be shorter than this. To get the test to complete in a reasonable time, we use shorter time between packets than usual as well as draw 0.1 mA between packets to drain the battery faster. We ran quite a lot of tests (39 batteries total) in order to make sure we had statistically valid results.
The most interesting thing, though, is that the effective battery capacity is the same for both the 15 mA load and the 30 mA load. The effective battery capacity turns out to be approximately 120 mAh (there is some variation battery to battery) for both 15 mA and 30 mA max loads using CR2032 cells at room temperature. Now, this means that the cell is used for pulsed operation, it must be de-rated from the 230 mAh that is typically given for constant current (usually given for a <1 mA load), but it does seems to disprove the theory that you cannot use coin cells if your (pulsed) load current is higher than 15 or 20 mA. All of these numbers are for a load without any form of decoupling capacitor used. Usually, you would use a moderately sized decoupling capacitor, which would increase the effective capacity by sourcing current during the current peak.
So it seems that at least in this case, the conventional wisdom does not seem to be based on hard facts.
BTW, we are planning to put together a full laboratory-style white paper when all our tests are complete, I'll let you know once that is available.
Food for thought,
Karl