# HG changeset patch # User Mychaela Falconia # Date 1622272025 0 # Node ID 1cdd0f0a6e700438b14189bca342e55a35f19675 # Parent ce044aa49bafbde62030aeeb7ab4821744ba5a5f Calypso-PWM-light and Calypso-buzzer-output articles written based on oscilloscope probing experiments diff -r ce044aa49baf -r 1cdd0f0a6e70 Calypso-PWM-light --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/Calypso-PWM-light Sat May 29 07:07:05 2021 +0000 @@ -0,0 +1,62 @@ +The Calypso chip has a PWM light output called LT/PWL - a digital output pin +that can be configured as either LT or PWL. The documentation we got from TI +is not exhaustively complete in describing the exact behaviour and output +waveforms of these two modes, thus I (Mother Mychaela) did a little bit of lab +experimentation to complete the picture. All experimental observations were +made with an oscilloscope probe placed on the LT/PWL signal on a FreeCalypso +Caramel2 board. + +LT mode +======= + +LT functionality of the LT/PWL output is implemented in the ARMIO block, +together with GPIO and keypad functions. LT output stays at constant low level +(zero light) only when the LIGHT bit in BUZZ_LIGHT_REG (FFFE:480E) is cleared; +if the LIGHT bit is set, then a small amount of light will be emitted even if +LIGHT_LEVEL_REG is set to 0, contrary to CAL207 document saying "no light" in +this case. Light levels 0 through 63 as written into LIGHT_LEVEL_REG really +correspond to 1/64 through 64/64 in terms of the actual emitted PWM duty cycles. +If the register is written with the maximum light level of 63, then LT output +is a continuous high level, with no o'scope-observable PWM activity. If the +register is written with any other value, then PWM activity becomes visible on +an oscilloscope, with each full cycle period equal to 64 periods of CLK13M, +i.e., PWM frequency equals 203.125 kHz, the master 13 MHz clock divided by 64. +The shape of this PWM output is totally straightforward: if LIGHT_LEVEL_REG is +set to 0, LT output is high for one CLK13M period and low for 63 periods, then +repeat; if LIGHT_LEVEL_REG is set to 1, LT output is high for 2 CLK13M periods +and low for 62 periods, and so forth, with LIGHT_LEVEL_REG set to 63 resulting +in LT being high in all 64 slots, i.e., continuous high output with no visible +PWM activity. + +PWL mode +======== + +The description of PWL in the CAL207 document is reasonably good; only a few +additional notes need to be made: + +* By the fundamental principles of how all LFSRs work, an LFSR of N bits CANNOT + have a period of 2**N, instead the greatest period that can be achieved with + careful choice of polynomial is 2**N-1. Calypso PWL block features an 8-bit + LFSR, TI's choice of polynomial (hard-wired in the silicon) is a proper one, + thus the period is 255. + +* Two alternate hw implementations are possible: XOR implementation would + produce an LFSR with valid values [1,255], whereas XNOR implementation would + produce an LFSR with valid values [0,254]. Obviously we have no way to look + inside Calypso silicon, but the visible behaviour with different comparator + values suggests that the internal LFSR runs in the [1,255] range. There must + be some reset logic in the hw that prevents a stuck value of 0. + +* The visible output on the PWL pin repeats every 255 cycles of CLK32K, and the + average light intensity ranges from 0/255 to 254/255 in 1/255 steps. If + PWL_LEVEL_REG is set to either 0 or 1, PWL output is continuous low; if the + register is set to 2, PWL output is 1/255 (on for just one CLK32K cycle out of + 255), and if the register is set to maximum of 255, PWL output is NOT + continuous high as CAL207 claims, but instead it is 254/255, i.e., on for 254 + out of every 255 CLK32K periods. + +Deep sleep interaction +====================== + +LT mode uses CLK13M, thus it is incompatible with deep sleep. PWL uses CLK32K +instead, thus deep sleep is perfectly OK with this light on. diff -r ce044aa49baf -r 1cdd0f0a6e70 Calypso-buzzer-output --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/Calypso-buzzer-output Sat May 29 07:07:05 2021 +0000 @@ -0,0 +1,91 @@ +Our dear Calypso has a dedicated digital output for driving old-fashioned +cellphone buzzers - I (Mother Mychaela) previously assumed those buzzers to be +piezoelectric, but now it appears that they are actually magnetic buzzers, not +piezo. But irrespective of the physics of the actual transducer that is +ultimately controlled by Calypso BU/PWT output, the Calypso output itself is +purely digital, producing digital waveforms, and we (FreeCalypso) need to +understand exactly what the chip puts out. The findings presented here have +been obtained by observing Calypso BU/PWT output with an oscilloscope; the hw +platform used by the Mother for these experiments was a BenQ M32 module - this +module has the familiar Calypso+Iota chipset inside, it has BU/PWT brought out, +and the breakout board supplied with these modules allowed for very quick +experimentation. + +BU mode +======= + +BU functionality of the BU/PWT output is implemented in the ARMIO block, +together with GPIO and keypad functions. Here are the new findings which were +previously unknowable without oscilloscope observation: + +* In "normal" buzzer operation when ARMIO_LOAD_TIM is in the [1,255] range + (i.e., not 0), the output frequency is 13 MHz / 512 / (ARMIO_LOAD_TIM + 1). + This part matches our previous understanding. The highest frequency that can + be produced is 12.6953125 kHz, and the lowest is 99.182129 Hz. + +* When ARMIO_LOAD_TIM is set to 0 and the buzzer is enabled, BU output is NOT a + 25.390625 kHz tone (per the formula above) as we previously thought - instead + BU output will be constant high if BUZZER_LEVEL_REG is set to maximum 63, or + pure 203.125 kHz PWM exactly like LT otherwise. + +* PWM power level control for BU works exactly like the one for LT - see the + companion Calypso-PWM-light article. + +* The interesting question is exactly how PWM power level control and tone + generation combine, and the answer turned out to match the terse description + in CAL207: BU is the output of an AND gate; one input to this AND gate is the + output of the timer that produces tones between 99 Hz and 12.7 kHz (or + constant high if ARMIO_LOAD_TIM is set to 0), and the other input to the AND + gate is a PWM block strictly identical to LT. + +* For a practical example, consider what happens when a 12.7 kHz (max frequency) + tone is emitted in combination with PWM. The "on" part of the waveform at + this tone frequency equals 512 periods of CLK13M, whereas the full cycle of + LT-like PWM is 64 periods of CLK13M. Thus the waveform seen on an o'scope + consists of 8 repeated PWM cycles, then a "pause" of 512 CLK13M periods (the + "off" part of the tone waveform), then the full cycle repeats. Furthermore, + the tone waveform and the PWM waveform fed to the internal AND gate are not + synchronized, and the resulting slight misalignment is easily visible on an + o'scope. + +Pirelli DP-L10 +============== + +In the Pirelli DP-L10 phone Calypso BU output is repurposed to control the +vibrator. Pirelli's official fw sets ARMIO_LOAD_TIM to 0 and BUZZER_LEVEL_REG +to 63; when operated in this manner, BU becomes a mere on/off output under +software control just like a GPIO. If someone wishes to run the vibrating motor +at a lower speed (i.e., exercise a form of "analog" control), the correct way to +perform such feat would be by dialing down BUZZER_LEVEL_REG (PWM control between +1/64 and 64/64) - setting ARMIO_LOAD_TIM to a nonzero value does not seem to +make any sense when the buzzer has been replaced with a vibrator. + +PWT mode +======== + +The principal difference between BU and PWT modes is that BU mode generates +"arbitrary" tone frequencies by dividing from CLK13M/512, whereas PWT mode +generates predefined musical note frequencies, specifically 48 notes of the +chromatic scale from F4 through E8 in the scientific pitch notation. The +frequency range of PWT mode is narrower than BU mode (349 Hz to 5274 Hz, as +opposed to 99 Hz to 12.7 kHz), and PWT does not support constant high output +for Pirelli-style vibrator driving - however, BU mode suffers from a major +problem in that its repertoire of possible tone frequencies does not correspond +to musical notes. Given the desire to have ringtone melodies that are composed +of standard musical notes, TI added the new PWT hardware block that is +specifically designed to produce musical note frequencies. + +Here are some observations about PWT made with o'scope experiments: + +* The description of FRC_REG in CAL207 is correct: bits [1:0] select the octave, + bits [5:2] select the semitone within the octave. The definitions in the + non-functional (not actually used) mmiBuzzer.h header file in TI's BMI source + are wrong - or more precisely, the definitions under #if 0 are the correct + ones, whereas the enabled-for-compilation definitions are wrong. + +* PWM power control works on the same principle as in BU mode: the output of + the musical tone frequency generator and the output of the PWM block are + simply ANDed together. The 64-level PWM block itself is very similar to the + ones in LT and BU, but it runs 8 times slower, i.e., one PWM period is 512 + cycles of CLK13M, as opposed to 64 cycles of CLK13M in BU and LT, and each + "quantum" of this slower PWM equals 8 cycles of CLK13M.