Post by thread:[PIC]: Ultra fast low res PWM challenge

I've spent some time playing with an RC car controller and needed a very fast 2 channel PWM routine. Resolution was not important, so 8 steps (3 bits) was fine. My first attempt which I'm sure can easily be bettered: clrf outport ; clear shadow register incf pwmcount,w ; increment pwm counter andlw 0x07 ; 8 step PWM, so roll counter over after 7 movwf pwmcount movf pwmset1,w subwf pwmcount,w ; subtract setpoint from counter skpc bsf outport,0 ; result negative so turn on PWM output in shadow register movf pwmset2,w subwf pwmcount,w ; subtract setpoint from counter skpc bsf outport,1 ; result negative so turn on PWM output in shadow register So this gives a 4 cycle overhead for the counter and clearing the shadow register, and 4 cycles for each additonal channel. This actually gives nine steps; 0 gives 0%, 8 or greater gives 100%. I have seen Scott Dattalos very fast PWM routines, but they seem to incur too much overhead if the duty cycle needs to be varied. Regards Mike -- http://www.piclist.com hint: To leave the PICList spam_OUTpiclist-unsubscribe-requestTakeThisOuTmitvma.mit.edu

I've spent some time playing with an RC car controller and needed a very fast 2 channel PWM routine. Resolution was not important, so 8 steps (3 bits) was fine. My first attempt which I'm sure can easily be bettered: clrf outport ; clear shadow register incf pwmcount,w ; increment pwm counter andlw 0x07 ; 8 step PWM, so roll counter over after 7 movwf pwmcount movf pwmset1,w subwf pwmcount,w ; subtract setpoint from counter skpc bsf outport,0 ; result negative so turn on PWM output in shadow register movf pwmset2,w subwf pwmcount,w ; subtract setpoint from counter skpc bsf outport,1 ; result negative so turn on PWM output in shadow register So this gives a 4 cycle overhead for the counter and clearing the shadow register, and 4 cycles for each additonal channel. This actually gives nine steps; 0 gives 0%, 8 or greater gives 100%, which is actually quite handy. I have seen Scott Dattalos very fast PWM routines, but they seem to incur too much overhead if the duty cycle needs to be varied. The pwm duty cyle is set every loop, so this must be taken into consideration. Regards Mike -- http://www.piclist.com hint: To leave the PICList .....piclist-unsubscribe-requestKILLspam@spam@mitvma.mit.edu

Hi I did something like this, ages ago (Z-80, 8085 era). I just use one byte with a circular counter. You just have to rotate and output one particular bit (B0 ie.). I couldn't find nothing faster than this, Z80 has rotates os shifts, I don't remember now, where the carry bit don't interfere with operation, it makes the things even easier ( b0->b7 , b1 ->b2,.... b7->b6). You just have to perform the rotation and output a particular bit. Hope it helps. Regards Horta {Quote hidden}> I've spent some time playing with an RC car controller and needed a very > fast 2 channel PWM routine. Resolution was not important, so 8 steps (3 > bits) was fine. > > My first attempt which I'm sure can easily be bettered: > > clrf outport ; clear shadow register > incf pwmcount,w ; increment pwm counter > andlw 0x07 ; 8 step PWM, so roll counter over after 7 > movwf pwmcount > > movf pwmset1,w > subwf pwmcount,w ; subtract setpoint from counter > skpc > bsf outport,0 ; result negative so turn on PWM output in > shadow register > > movf pwmset2,w > subwf pwmcount,w ; subtract setpoint from counter > skpc > bsf outport,1 ; result negative so turn on PWM output in > shadow register > > So this gives a 4 cycle overhead for the counter and clearing the shadow > register, and 4 cycles for each additonal channel. This actually gives nine {Quote hidden}> steps; 0 gives 0%, 8 or greater gives 100%. > > I have seen Scott Dattalos very fast PWM routines, but they seem to incur > too much overhead if the duty cycle needs to be varied. > > Regards > > Mike > > -- > http://www.piclist.com hint: To leave the PICList > piclist-unsubscribe-requestKILLspammitvma.mit.edu > > > -- http://www.piclist.com hint: PICList Posts must start with ONE topic: [PIC]:,[SX]:,[AVR]: ->uP ONLY! [EE]:,[OT]: ->Other [BUY]:,[AD]: ->Ads

{Quote hidden}> -----Original Message----- > From: PY2NI [SMTP:.....py2niKILLspam.....UOL.COM.BR] > Sent: Thursday, February 21, 2002 12:10 PM > To: EraseMEPICLISTspam_OUTTakeThisOuTMITVMA.MIT.EDU > Subject: Re: [PIC]: Ultra fast low res PWM challenge > > Hi I did something like this, ages ago (Z-80, 8085 era). I just use > one > byte with a circular counter. You just have to rotate and output one > particular bit (B0 ie.). I couldn't find nothing faster than this, Z80 has > rotates os shifts, I don't remember now, where the carry bit don't > interfere > with operation, it makes the things even easier ( b0->b7 , b1 ->b2,.... > b7->b6). > You just have to perform the rotation and output a particular bit. > Hope > it helps. > > > Regards > Horta > Thanks for the tip, by coincidence I coded a rouine that does exactly that last night. clrf outport ; clear all output bits in the shadow register rrf pwm1,w ; put bit 0 into carry rrf pwm1,f ; rotate byte skpnc bsf outport,0 ; if carry set and set output bit rrf pwm2,w ; put bit 0 into carry rrf pwm2,f ; rotate byte skpnc bsf outport,1 ; if carry set, then set output bit movf outport ; update the PORT from the shadow reg movwf PORTB This is actually no faster than the original method, but has a couple of advantages, and one disadvantage. The disadvantage is that the duty cycle can no longer be set directly, you have to translate to a bit pattern which adds some overhead: setpwm: ; call with duty cycle in W. Need to add any bank switching, and limit maximum W!!! addwf PCL,f retlw b'00000000' retlw b'00000001' retlw b'00010001' retlw b'01001001' retlw b'01010101' retlw b'10110101' retlw b'11101101' retlw b'01111111' retlw b'11111111' To set a pwm channel you'd simply use the following which (I think) takes 8 cycles. movlw 5 call setpwm movwf pwm1 One advantage of this method is that one byte less RAM is needed (no PWM base counter), the other (possible) advantage is that the PWM frequency for mid range settings can be increased which can make filtering easier, or maybe reduce motor whine. This is quite a nice way to do things if you have the PWM running under an interrupt timer and it is not updated very frequently. Regards Mike -- http://www.piclist.com hint: PICList Posts must start with ONE topic: [PIC]:,[SX]:,[AVR]: ->uP ONLY! [EE]:,[OT]: ->Other [BUY]:,[AD]: ->Ads

If you can live with 256 steps of resolution then you can use something like: clrf Fref incf Fref,f ; Fref = 1 pwm_again: incfsz Fref,f goto ($+2) clrf Fio ; all channels are 0 ; first channel movlw CH0_MASK ; has a 1 in the o/p of that mask incfsz Fch0,f ; increment channel data directly goto ($+2) iorwf Fio,f ; when crossing 0 set the bit ; next channel movlw CH1_MASK incfsz Fch1,f goto ($+2) iorwf Fio,f ; ... more like this ; code to load etc channels here (your choice) goto pwm_again ; forever The code has constant run time and is very flexible, for example you can replace the iorwf with xorwf or you can implement full stroke (differential) PWM to achieve +/-5v excursion by using two io pins for a channel and activating them 180 degrees out of phase, f.ex. by replacing the clrf at the top of the loop with a suitable load and the iorwf of th channel with a xorwf so both bits toggle together. There will be a small bias (offset) added to each channel because of its position in the loop (wrt the clrf). If you allow the loop to run 256 times then you do not need to reload the pwm counters in despite of their incrementing. You can also sequence the channels (must if your PSU is crummy) by replacing the incfsz with decfsz on certain channels. The Fref contains a running phase reference that can be useful. The loop timing is Tloop = 4 + 5 * channels + Your_Overhead + 2 cycles. This times 256 is the PWM frequency. For 20 msec PWM period (standard servos) and 4MHz clock and no interrupts (bit banged) you have 'time' for 14 channels in the loop. You can also eliminate the loop and have a fast timer ISR run the code (but probably not at 4MHz). If you reload the channel data at an unfortunate time the respective output may skip one output pulse. To avoid this reload when Fref == 1 at the bottom of the loop (or 0 at top). hope this helps, Peter -- http://www.piclist.com hint: PICList Posts must start with ONE topic: [PIC]:,[SX]:,[AVR]: ->uP ONLY! [EE]:,[OT]: ->Other [BUY]:,[AD]: ->Ads