mirror of
https://github.com/TomHarte/CLK.git
synced 2024-11-17 10:06:21 +00:00
80 lines
3.4 KiB
C++
80 lines
3.4 KiB
C++
//
|
|
// DriveSpeedAccumulator.cpp
|
|
// Clock Signal
|
|
//
|
|
// Created by Thomas Harte on 01/06/2019.
|
|
// Copyright © 2019 Thomas Harte. All rights reserved.
|
|
//
|
|
|
|
#include "DriveSpeedAccumulator.hpp"
|
|
|
|
namespace {
|
|
|
|
/*
|
|
For knowledge encapsulate below, all credit goes to the MAME team. No original research here.
|
|
|
|
Per their investigation, the bytes collected for PWM output feed a 6-bit LFSR, which then keeps
|
|
output high until it eventually reaches a state of 0x20. The LFSR shifts rightward and taps bits
|
|
0 and 1 as the new input into bit 5.
|
|
|
|
I've therefore implemented the LFSR as below, feeding into a lookup table to calculate actual
|
|
pulse widths from the values stored into the PWM buffer.
|
|
*/
|
|
template<uint8_t value> constexpr uint8_t lfsr() {
|
|
if constexpr (value == 0x20 || !value) return 0;
|
|
return 1+lfsr<(((value ^ (value >> 1))&1) << 5) | (value >> 1)>();
|
|
}
|
|
|
|
constexpr uint8_t pwm_lookup[] = {
|
|
lfsr<0>(), lfsr<1>(), lfsr<2>(), lfsr<3>(), lfsr<4>(), lfsr<5>(), lfsr<6>(), lfsr<7>(),
|
|
lfsr<8>(), lfsr<9>(), lfsr<10>(), lfsr<11>(), lfsr<12>(), lfsr<13>(), lfsr<14>(), lfsr<15>(),
|
|
lfsr<16>(), lfsr<17>(), lfsr<18>(), lfsr<19>(), lfsr<20>(), lfsr<21>(), lfsr<22>(), lfsr<23>(),
|
|
lfsr<24>(), lfsr<25>(), lfsr<26>(), lfsr<27>(), lfsr<28>(), lfsr<29>(), lfsr<30>(), lfsr<31>(),
|
|
lfsr<32>(), lfsr<33>(), lfsr<34>(), lfsr<35>(), lfsr<36>(), lfsr<37>(), lfsr<38>(), lfsr<39>(),
|
|
lfsr<40>(), lfsr<41>(), lfsr<42>(), lfsr<43>(), lfsr<44>(), lfsr<45>(), lfsr<46>(), lfsr<47>(),
|
|
lfsr<48>(), lfsr<49>(), lfsr<50>(), lfsr<51>(), lfsr<52>(), lfsr<53>(), lfsr<54>(), lfsr<55>(),
|
|
lfsr<56>(), lfsr<57>(), lfsr<58>(), lfsr<59>(), lfsr<60>(), lfsr<61>(), lfsr<62>(), lfsr<63>(),
|
|
};
|
|
|
|
}
|
|
|
|
using namespace Apple::Macintosh;
|
|
|
|
void DriveSpeedAccumulator::post_sample(uint8_t sample) {
|
|
if(!delegate_) return;
|
|
|
|
// An Euler-esque approximation is used here: just collect all
|
|
// the samples until there is a certain small quantity of them,
|
|
// then produce a new estimate of rotation speed and start the
|
|
// buffer afresh.
|
|
//
|
|
// Note the table lookup here; see text above.
|
|
sample_total_ += pwm_lookup[sample & 0x3f];
|
|
++sample_count_;
|
|
|
|
if(sample_count_ == samples_per_bucket) {
|
|
// The below fits for a function like `a + bc`; it encapsultes the following
|
|
// beliefs:
|
|
//
|
|
// (i) motor speed is proportional to voltage supplied;
|
|
// (ii) with pulse-width modulation it's therefore proportional to the duty cycle;
|
|
// (iii) the Mac pulse-width modulates whatever it reads from the disk speed buffer, as per the LFSR rules above;
|
|
// (iv) ... subject to software pulse-width modulation of that pulse-width modulation.
|
|
//
|
|
// So, I believe current motor speed is proportional to a low-pass filtering of
|
|
// the speed buffer. Which I've implemented very coarsely via 'large' bucketed-averages,
|
|
// noting also that exact disk motor speed is always a little approximate.
|
|
|
|
// The formula below was derived from observing values the Mac wrote into its
|
|
// disk-speed buffer. Given that it runs a calibration loop before doing so,
|
|
// I cannot guarantee the accuracy of these numbers beyond being within the
|
|
// range that the computer would accept.
|
|
const float normalised_sum = float(sample_total_) / float(samples_per_bucket);
|
|
const float rotation_speed = (normalised_sum - 3.7f) * 17.6f;
|
|
|
|
delegate_->drive_speed_accumulator_set_drive_speed(this, rotation_speed);
|
|
sample_count_ = 0;
|
|
sample_total_ = 0;
|
|
}
|
|
}
|