cac_accuracy_demo Directory Reference

Directory dependency graph for cac_accuracy_demo:

Directories
	src

Detailed Description

Uses the Clock Frequency Accuracy Measurement Circuit to count edges of the 24 MHz main oscillator inside one window of the LOCO reference divided by 32, and flags a frequency error if the crystal is off. LED1 toggles on a healthy measurement and LED2 on a timeout or error; g_cac_count, g_cac_ok, g_cac_status and g_cac_heartbeat mirror the result for a headless probe.

The allowed window is the expected count (24 MHz over the 1024 Hz reference, about 23437) plus or minus roughly 6 percent – wide enough that a healthy crystal never trips FERRF, narrow enough that a stopped or grossly detuned oscillator always does. The LOCO must be running: it is the reference, and if it is stopped the measurement simply times out rather than reporting an error about the crystal. No external hardware.

The ra8_cac HAL takes only the count limits and leaves CACR1 / CACR2 at zero, which is a degenerate main-against-main pairing. This demo writes those two registers directly to select the real main-against-LOCO pair; folding clock-source selection into ra8_cac_init would remove the need.

Configuration, HUM R01UH1065EJ0130 Ch 10 "CAC":

• CAULVR / CALLVR carry the window (Ch 10.2.6 / 10.2.7 p 425).
• CACR1 = 0x00 – target is the main oscillator (FMCS = 000), no division, rising edge (Ch 10.2.2 "CACR1" p 421).
• CACR2 = 0x09 – internal reference (RPS = 1), LOCO (RSCS = 100) divided by 32 (RCDS = 00), no digital filter (Ch 10.2.3 "CACR2" p 422).
• A measurement sets CACR0.CFME and polls CASTR.MENDF (Ch 10.2.1 p 421, Ch 10.2.5 p 424); the count is read from CACNTBR.

The emulator models the edge counter by loading CACNTBR with the midpoint of whatever window the firmware programmed, so it is in band by construction. That proves the start / poll / read-back sequence and nothing whatever about the real clock; only silicon proves the edge count.