An external 32 kHz crystal will give lower average current consumption than using the RC oscillator, at the expense of cost and board space. There are primarily 3 reasons the current consumption is lower:
The RC oscillator must be calibrated every 4 seconds to be accurate enough. Calibration normally takes about 17 ms, and requires the 16 MHz to be running, giving an average current consumption increase of about 6-7 µA.
The receive window must be widened to account for the worse accuracy. Since the RC is 250 ppm, compared with typically 20-30 ppm for an external crystal, the radio must receive for a slightly longer time, giving an increase of 1-2 µA average
The run current or the RC is 0.8 µA instead of 0.4 µA for a typical crystal.
In total, this gives 8-10 µA extra average current consumption. Depending on application, this may be enough to justify the cost and board space for a crystal.
The reason I ask is the we are not currently using an external 32KHz crystal and we power down the chip externally when we are not expecting BLE activity. So the calibration you speak of can never happen. Is this OK?
The calibration is only needed as long as the chip is actually running, so the calibration will still run when needed, no matter if you power off the chip externally.
If you take a look in the nRF51822 PS, you can see that it is specified to be less that 250 ppm when calibrated every 4 seconds as long as the temperature is constant. This is within the requirements given in the Bluetooth Core Specification.
