"MTX Plus⁺" Firmware

 

CPU Board

System Clock Generator

Tony Brewer designed the CPLD / GAL logic that divides the input clock source to generate a range of frequencies that can be used to clock the CPU, the system can dynamically switch between the various clock speeds using the system control I/O port (255 / FF_h). Tony has written some notes that explain how the Clock Generator works . . . . .

A 3-bit counter (outputs Q0, Q1, Q2) divides the 32 MHz master clock by between two and eight. There are four count sequences as follows:   01234567 -  divide by 8 and 4 and 2 0124567 -  divide by 7 and 3½ 012456 -  divide by 6 and 3 01245 -  divide by 5 and 2½ The nine clocks listed above are displayed below, together with three times divide by 7 or 96/7 = 13.71 MHz (not used). Where there is more than one option for the same frequency the phase, duty cycle or minimization of logic are considered. The nine chosen options are marked with an asterisk.
	Count	Divide	Frequency
	0 1 2 3 4 5 6 7	by	(MHz)	Duty Cycle
Q0		2	16.0 *	50 : 50
Q1		4	8.0 *	50 : 50
Q2		8	4.0 *	50 : 50
Q0		2.33	13.71	50 : 50 / 50 : 50 / 67 : 33
Q1		3½	9.14	50 : 50 / 67 : 33
Q2		7	4.57	57 : 43
Q0		2.33	13.71	50 : 50 / 50 : 50 / 67 : 33
Q1		3½	9.14 *	50 : 50  / 67 : 33
Q2		7	4.57 *	57 : 43
Q0		3	10.67	67 : 33
Q1		3	10.67 *	67 : 33
Q2		6	5.33 *	50 : 50
Q0		2½	12.8 *	50 : 50 / 67 : 33
Q1		5	6.4	80 : 20
Q2		5	6.4 *	60 : 40
Q0		2½	12.8	50 : 50 / 67 : 33
Q1		5	6.4	80 : 20
Q2		5	6.4	60 : 40
Additional clocks can be created by combining two of the nine clocks, for example alternating /6 and /7 produces /6½. Seven such alternating clocks have been added to the nine single-frequency clocks to give a total of 16 as follows: System PHI (Clock Frequency 32MHz) Clock D7 D6 D5 D4 Decimal Divisor PHI Q? 0 0 0 0 0 0 /8.00 4.00 MHz Q2 1 0 0 0 1 16 /7.00 4.57 MHz Q2 2 0 0 1 0 32 /6.50 4.92 MHz Q2-Q2 3 0 0 1 1 48 /6.00 5.33 MHz Q2 4 0 1 0 0 64 /5.50 5.82 MHz Q2-Q2 5 0 1 0 1 80 /5.00 6.40 MHz Q2 6 0 1 1 0 96 /4.50 7.11 MHz Q2-Q1* 7 0 1 1 1 112 /4.00 8.00 MHz Q1 8 1 0 0 0 128 /3.75 8.53 MHz Q1-Q1 9 1 0 0 1 144 /3.50 9.14 MHz Q1 10 1 0 1 0 160 /3.25 9.85 MHz Q1-Q1 11 1 0 1 1 176 /3.00 10.67 MHz Q1 12 1 1 0 0 192 /2.75 11.64 MHz Q1-Q0 13 1 1 0 1 208 /2.50 12.80 MHz Q0 14 1 1 1 0 224 /2.25 14.22 MHz Q1-Q0* 15 1 1 1 1 240 /2.00 16.00 MHz Q0 Each option selects a unique clock, apart from 0, the odd clocks are single-frequency and the even ones alternating. Clocks 9-15 are twice the speed of the clocks eight lower and bits D (6:5) specify the count sequence (00 for /7, 01 for /6, 10 for /5 and 11 for /8) for eight of the nine single-frequency clocks, both of which help to reduce the logic required greatly. The Q? column shows the Q output used for PHI, inverted and delayed by 1T at 32 MHz. Five alternating clocks have the same Q output and two different but it makes no difference to the PHI logic. Clocks 6 and 14 require special treatment and are marked with an asterisk. A 1-bit counter selects which of the two count sequences should be used for an alternating clock and is toggled when the count is zero. However, clock 7 (/4) contains two periods between zero counts whereas clock 5 (/5) has one and clock 15 (/2) has four periods but clock 13 (/2½) only two. Alternating the clocks equally will not generate the correct frequencies for clocks 6 and 14. There are two solutions to this problem. The first uses a 2-bit counter that counts 012012... for these two clocks and 01230123... for the rest. Bit 0 of the counter will be 010010... for the special cases and 01010101... for the others and is used to select the lower frequency clock when 0, thus two periods of /5 are followed by two of /4 for example. The second, and better, solution is to change the count sequence for /4 and /2 from 01234567 to 0123 so that there is only one period of /4 and two of /2 between zero counts, the same as for /5 and /2½, respectively. Both options have been tested successfully in the CPLD. The second requires less logic and has an alternating period half as long as the first for clocks 6 and 14. Five of the 16 clocks can generate a 125 Hz interrupt precisely and the rest have small errors as listed below in the table of CTC time constants, with more than one pair of values possible for most clocks. Clock /? MHz 125 Hz Error CTC Time Constants 0 8 4.00    16,125 (4,000,000) 1 7 4.57  1s in 8001s 18,127 (4,572,000) 2 6½ 4.92  1s in 4570s 23,107 (4,922,000) 3 6 5.33  1s in 8001s 21,127 (5,334,000) 4 5½ 5.82  1s in 3201s 15,194 (5,820,000) 5 5 6.40    25,128 (6,400,000) 6 4½ 7.11  1s in 8001s 28,127 (7,112,000) 7 4 8.00    32,125 (8,000,000) 8 3¾ 8.53  1s in 12801s 17,251 (8,534,000) 9 3½ 9.14  1s in 8001s 36,127 (9,144,000) 10 3¼ 9.85  1s in 4570s 46,107 (9,844,000) 11 3 10.67 1s in 8001s 42,127 (10,668,000) 12 2¾ 11.64 1s in 7112s 23,253 (11,638,000) 13 2½ 12.80   50,128 (12,800,000) 14 2¼ 14.22 1s in 80011s 56,127 (14,224,000) 15 2 16.00   64,125 (16,000,000)