LFO odd dividers, triplets and others.

[ganchan]

i'm trying to make odd & new values in my LFOs. like 3/1 9/1 or 6/1 12/1 etc. and something like 1/2T 1/4T etc. T=triplets.

the problem is the divider of 1024 that report comma values and i've not completly understand how work the sync but i've make some try just testing it hearing and counting. not so pro.

some other mathematic info would be cool.

i've look at

SIDSE_M_Sync

SIDSE_M_LFO_Restart

SID_LFO_TABLE_MCLK

tonight i will post some other info, i'm @ wrk now.  :)

[stryd_one]

Heya,

Yeh, floating point won't work ;) The best way to do this if the values will be static (like yours), is with a table of the results of the division, rounded to the nearest integer.

However you won't need to do that for the standard measures like 1/3,1/12,4/3,3/8, etc, that's why midi clock is 24ppqn, its the lowest common multiple of all of the common figures. MMA were thinking about us heheheh

[ganchan]

yes i've thought about the nearest value. in ex 3/1 it's 341.33333333333333333333.

so i can use 341 instead, the difference it's minimal. so the wave come out of the point after a lot of loops.

but i think to a function (to now it's just an idea, no code wrote down), that restart the lfo after a given time/loop where the wave would be start in sync with the seq. so i can pseudo preserve the loop point of LFO start.

[/tilted/]

i'm trying to make odd & new values in my LFOs. like 3/1 9/1 or 6/1 12/1 etc. and something like 1/2T 1/4T etc. T=triplets.

the problem is the divider of 1024 that report comma values and i've not completly understand how work

I of course have no idea here, but:

Is it possible to change the divider?

To maybe 1536 or 768?

Both significant numbers in the binary world...

unfortunately the only way to ensure even divisions regardless of your time sig is to use ? ppqn.

This makes your sequencer run very slowly...

Edit: fixed the ?...

[ganchan]

yes if 1/1 was 1536 in the sync table there was no problem. since midiclock it's 24ppqn, we must wrote a parallel counter that for each clock use 1536 in sync, and the lfo table it's referred to this.

[stryd_one]

yes i've thought about the nearest value. in ex 3/1 it's 341.33333333333333333333.

so i can use 341 instead, the difference it's minimal. so the wave come out of the point after a lot of loops.

Yeh, that would suck... That's why that's not what I mean ;)

I find the best way to avoid floating point is to avoid floating point hehehe... Just stay with good old fashioned fractions

1024/3=341 and 1/3

or you could say

1024/3=341 with a modulus (remainder) of 1

So can use 341, but if you want it to stay in sync, you've gotta spread the modulus around. You can use calculations to do it, but that could be a bit too much for the PIC to handle at such high frequency.

However if you know what values you want to have available (like how you have 3/1), a nice quick way to do it is a lookup table, which has the pre-calculated period of each tick in it like so:

341

342

341

Tada, synced clocks. You just look up the value according to how many times it's ticked, and roll that counter over every 3 ticks. But when it comes to fast stuff like LFO's... You want to be careful how you time it... Try to look up the value pre-emptively.

Notice that I put the remainder in the middle... That's where it really goes, because you want to just round to the nearest integer to be really accurate...it's like

0

+ 341 1/3 = 341 1/3 ~ 341 difference from the last is 341

+ 341 1/3 = 682 2/3 ~ 683 difference from the last is 342 there's our modulus

+ 341 1/3 = 1024 ~ 1024 difference from the last is 341

Dunno how much of that you'll be able to roll into the app though....

HTH!

Edit: I just saw your mention of the SID LFOs... if you're thinking about doing this in the MBSID, I'm not sure it'll work... I need to build one and figure out how the LFO's work :)

[/tilted/]

... But when it comes to fast stuff like LFO's...

You crazy kids and your fast LFOs.

I remember when the 'L' was for "low"  ;)

Mind you, according to the electricians, 240V household mains is referred to as "low voltage".

They count anything less than 1000V as "low".

-cause they're tough.  :)

[ganchan]

this sound very interesting and yes you have figured right about my skill to do that but how knows.

so you mean a table where i take for each LFO period a value in ex 314, than 342, than 341 and so on.

so for each divider i must have 3 tick cycled by the function.

for now this is my temporary new table for the bassline:

SID_LFO_TABLE_MCLK
	dw	16		; 231: 64/1
	dw	32		; 232: 32/1
	dw	42		; 233: 24/1 (42.66666666666666666667)
	dw	57		; 234: 18/1 (56.88888888888888888889)
	dw	64		; 235: 16/1
	dw	85		; 236: 12/1 (85.33333333333333333333)
	dw	114		; 237: 9/1 (113.77777777777777777778)
	dw	128		; 238: 8/1
	dw	170		; 239: 6/1 (170.66666666666666666667)
	dw	256		; 240: 4/1
	dw	341		; 241: 3/1 (341.33333333333333333333)
	dw	512		; 242: 2/1
	dw	614		; 243: 5/3 (614.4)
	dw	768		; 244: 4/3
	dw	1024		; 245: 1/1
	dw	1365		; 246: 3/4 (1365.33333333333333333333)
	dw	1536		; 247: 2/3
	dw	1638		; 248: 5/8 (1638.4)
	dw	2048		; 249: 1/2
	dw	2730		; 250: 3/8 (2730.66666666666666666667)
	dw	3072		; 251: 1/3
	dw	4096		; 252: 1/4
	dw	8192		; 253: 1/8
	dw	12288		; 254: 1/12
	dw	16384		; 255: 1/16

CS_MENU_PRINT_LFOx_RATE
	rcall	CS_MENU_PRINT_GetLFOx	; FSR0 will point to SID_Ix_LFOx_RATE

	;; prepare TBLPTR[LH]
	TABLE_ADDR CS_MENU_PRINT_LFOx_RATE_Tab
	;; if clock sync activated, and rate >= 245, print period length
	movlw	(SID_Ix_LFOx_MODE-SID_Ix_LFOx_RATE)
	BRA_IFCLR PLUSW0, SID_I_LFO_MODE_CLKSYNC, ACCESS, CS_MENU_PRINT_LFOx_RATE_Dec
	movf	INDF0, W
	addlw	-231
	bnc	CS_MENU_PRINT_LFOx_RATE_Dec

SIDSE_M_LFO_Rate_ClkSync
	movf	IRQ_TMP1, W
	addlw	-231
	bnc	SIDSE_M_LFO_Rate_NoClkSyncValues
	TABLE_ADDR_MUL_W SID_LFO_TABLE_MCLK, 2
	rgoto	SIDSE_M_LFO_Rate_ClkSync_Cont

CS_MENU_PRINT_LFOx_RATE_Tab
	dw	"64/1"
	dw	"32/1"
	dw	"24/1"
	dw	"18/1"
	dw	"16/1"
	dw	"12/1"
	dw	"9/1 "
	dw	"8/1 "
	dw	"6/1 "
	dw	"4/1 "
	dw	"3/1 "
	dw	"2/1 "
	dw	"5/3 "
	dw	"4/3 "
	dw	"1/1 "
	dw	"3/4 "
	dw	"2/3 "
	dw	"5/8 "
	dw	"1/2 "
	dw	"3/8 "
	dw	"1/3 "
	dw	"1/4 "
	dw	"1/8 "
	dw	"1/12"
	dw	"1/16"

the slow lfo curve it's my point, 9/1 and 8/1 in ex make some cool modulation on 2 cloned bassline in sync.

-6/1 and 3/1 too

[stryd_one]

the slow lfo curve it's my point, 9/1 and 8/1 in ex make some cool modulation on 2 cloned bassline in sync.

Now you know why I'm all about polyrhythms ;)

[TK.]

In MBSEQ, triplets are generated by selecting a different predivider.

The internal 96 ppqn are divided by 6 (-> 16 ppqn) and 4 (-> 24 ppqn) to avoid fractions or expensive math

There are two reasons why I haven't done this for MBSID yet. One reason is, that SID_SE_STATE (which propagates the predivided clock events... and more flags) is fully allocated, another reason is, that SID_BASE section 0x100..0x1ff is completely allocated as well, so that I need to find some ways to free memory before introducing such special features.

It's still in my focus to get the firmware bugfree before starting with optimisations, a proper solution for triplets will be provided by default in a pre 2.0 release.

Of course, meanwhile you are free to tinker on your own solution, or to remove features you don't need (variables in SID_BASE section) to get some free memory...

Best Regards, Thorsten.

[ganchan]

tnx for the info, i will go to see MBSEQ functions!

since i need only the bassline engine in my Datassette i will cut here and there some code. :)

[stryd_one]

Did you miss this in my first post ganchan?

[quote name=stryd_one link=topic=11311.msg89340#msg89340 date=1209486777

However you won't need to do that for the standard measures like 1/3,1/12,4/3,3/8, etc, that's why midi clock is 24ppqn, its the lowest common multiple of all of the common figures. MMA were thinking about us heheheh

[ganchan]

no stryd i've not miss! i've just take a look at the MBSEQ code to improve my skill and learn about clock dividers. i'm not able at now to do that, i'm still on the way, just studying and tinkering with the support of my brother (informatic engineer), and your obviously.  :)

[stryd_one]

Ahh no worries man :) You might like to check out the enhanced clockbox too, to see the dividers at work in C.

[ganchan]

well, no good news about my code, i've make some try nothing working for now :p, so here i've attached a file that can show a little (not so good for real) what can do odd lfo long periods.

2 BassLine Seq. lines, 2 notes each, other notes are glided.

Lfo1 L PTr 4/3 DPW115 SncNo

Lfo2 L PSw 3/1 D.F40 SncYes

Lfo1 R off

Lfo2 R PWs 3/4 D.F30 SncNo

Odd_Lfo_Values.mp3

Odd_Lfo_Values.mp3