AN934

Calculating the needed amount of calibration

Doc ID 6393 Rev 4

7/14

Calculating the needed amount of calibration

There are two methods for establishing how much calibration are required in a given

application.

The first method can be easily implemented in the user environment simply by setting the

clock to a known accurate reference and then storing the time in some unused location in

the RAM. Over a period of time (30 days), the reference time is compared to the current time

(the average ambient temperature should be considered as well). This will tell the user how

fast or slow the clock operates for a period of 30 days. While it may seem crude, it allows the

designer to give the end user the ability to calibrate the clock according to the specified

environment. The ability to calibrate the clock can also be accomplished even after the final

product is packaged in a non-user serviceable enclosure by providing a simple utility to

access the calibration byte.

Table 1 on page 6 provides a direct look-up table for calibration values based upon the

number of seconds lost or gained during a one month (30 day) period. For example, if the

system were to lose 20 seconds during one month, the needed calibration would be +20

seconds. The user could look up a +20 second value in the table under the appropriate

column. In this case, the nearest value is +21. The appropriate sign bit in this case is a

logical '1,' indicating the clock needs to speed up to compensate for the lost time. This yields

a calibration value of “100010.” In this example, the inaccuracy would be reduced from –20

seconds per month to +1 second per month.

The second approach is better suited for a manufacturing environment, and involves the use

of a special test mode (as described in the section entitled, “Enabling the frequency test

function (FT)” which derives a 512Hz signal from the clock divider chain, as indicated in

Figure 2. This signal can be used to measure the accuracy of the crystal oscillator. The

right-hand pair of columns in Table 1 on page 6 provides a look-up table similar to that in the

left-hand pair of columns, except that the error values are expressed in “ppm” units instead

of seconds per month. The error in ppm can be quickly calculated by dividing the measured

error from 512 Hz by 512 and multiplying the result by 1 million. For example, if the

frequency measured during the test mode is 511.998 Hz, the delta is –0.002. Dividing by

512 and multiplying by 1 million, the result is –3.906 ppm. In this case, the nearest

compensation value is a +4.068. The appropriate sign bit in this case is a logical '1,'

indicating the clock needs to speed up to compensate for the lost time. This yields a

calibration value of “100001.”