UHSDR/UHSDR-active-devel/mchf-eclipse/drivers/audio/cw/cw_decoder.c
2022-11-08 16:13:55 +01:00

1109 lines
37 KiB
C
Raw Blame History

/************************************************************************************
** **
** UHSDR Firmware Project **
** **
**---------------------------------------------------------------------------------**
** **
** Licence: GNU GPLv3 **
************************************************************************************/
//*********************************************************************************
//**
//** Project.........: Read Hand Sent Morse Code (tolerant of considerable jitter)
//**
//** Copyright (c) 2016 Loftur E. Jonasson (tf3lj [at] arrl [dot] net)
//**
//** This program is free software: you can redistribute it and/or modify
//** it under the terms of the GNU General Public License as published by
//** the Free Software Foundation, either version 3 of the License, or
//** (at your option) any later version.
//**
//** This program is distributed in the hope that it will be useful,
//** but WITHOUT ANY WARRANTY; without even the implied warranty of
//** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
//** GNU General Public License for more details.
//**
//** The GNU General Public License is available at
//** http://www.gnu.org/licenses/
//**
//** Substantive portions of the methodology used here to decode Morse Code are found in:
//**
//** "MACHINE RECOGNITION OF HAND-SENT MORSE CODE USING THE PDP-12 COMPUTER"
//** by Joel Arthur Guenther, Air Force Institute of Technology,
//** Wright-Patterson Air Force Base, Ohio
//** December 1973
//** http://www.dtic.mil/dtic/tr/fulltext/u2/786492.pdf
//**
//** Platform........: Teensy 3.1 / 3.2 and the Teensy Audio Shield
//**
//** Initial version.: 0.00, 2016-01-25 Loftur Jonasson, TF3LJ / VE2LJX
//**
//*********************************************************************************
#include "uhsdr_board.h"
#include "ui_lcd_layouts.h"
#include "ui_driver.h"
#include "cw_decoder.h"
#include "audio_driver.h"
#include "rtty.h"
#include "cw_gen.h"
#include <stdio.h>
Goertzel cw_goertzel;
cw_config_t cw_decoder_config =
{ .sampling_freq = 12000.0, .target_freq = 750.0,
.speed = 25,
// .average = 2,
.thresh = 32000,
.blocksize = CW_DECODER_BLOCKSIZE_DEFAULT,
// .AGC_enable = 0,
.noisecancel_enable = 1,
.spikecancel = 0,
.use_3_goertzels = false,
// .snap_enable = true,
.snap_enable = false,
.show_CW_LED = true, // menu choice whether the user wants the CW LED indicator to be working or not
};
static void CW_Decode(void);
void CwDecode_Filter_Set()
{
// set Goertzel parameters for CW decoding
AudioFilter_CalcGoertzel(&cw_goertzel, ts.cw_sidetone_freq , // cw_decoder_config.target_freq,
cw_decoder_config.blocksize, 1.0, cw_decoder_config.sampling_freq);
}
//#define SIGNAL_TAU 0.01
#define SIGNAL_TAU 0.1
#define ONEM_SIGNAL_TAU (1.0 - SIGNAL_TAU)
#define CW_TIMEOUT 3 // Time, in seconds, to trigger display of last Character received
#define ONE_SECOND (12000 / cw_decoder_config.blocksize) // sample rate / decimation rate / block size
//#define CW_ONE_BIT_SAMPLE_COUNT (ONE_SECOND / 5.83) // standard word PARIS has 14 pulses & 14 spaces, assumed: 25WPM
#define CW_ONE_BIT_SAMPLE_COUNT (ONE_SECOND / 58.3) // = 6.4 works ! standard word PARIS has 14 pulses & 14 spaces, assumed: 25WPM
//#define CW_ONE_BIT_SAMPLE_COUNT (ONE_SECOND / 25.0) // does not work! standard word PARIS has 14 pulses & 14 spaces, assumed: 25WPM
//#define CW_ONE_BIT_SAMPLE_COUNT (ONE_SECOND / 583.0) // does not work. standard word PARIS has 14 pulses & 14 spaces, assumed: 25WPM
//#define CW_ONE_BIT_SAMPLE_COUNT (12000 / 5.83) // standard word PARIS has 14 pulses & 14 spaces, assumed: 25WPM
//#define CW_ONE_BIT_SAMPLE_COUNT (12000) // standard word PARIS has 14 pulses & 14 spaces, assumed: 25WPM
// 14bits * 25words per min / (60 sec/min) = 5.83 bits/sec bitrate
// (sample_rate / blocksize) / bitrate = samples per bit !
#define CW_SPIKECANCEL_MAX_DURATION 8 // Cancel transients/spikes/drops that have max duration of number chosen.
// Typically 4 or 8 to select at time periods of 4 or 8 times 2.9ms.
// 0 to deselect.
#define CW_SIG_BUFSIZE 256 // Size of a circular buffer of decoded input levels and durations
#define CW_DATA_BUFSIZE 40 // Size of a buffer of accumulated dot/dash information. Max is DATA_BUFSIZE-2
// Needs to be significantly longer than longest symbol 'sos'= ~30.
// FIXME: replace with true/false (since already defined elsewhere)
#define TRUE 1
#define FALSE 0
typedef struct
{
unsigned state :1; // Pulse or space (sample buffer) OR Dot or Dash (data buffer)
unsigned time :31; // Time duration
} sigbuf;
typedef struct
{
unsigned initialized :1; // Do we have valid time duration measurements?
unsigned dash :1; // Dash flag
unsigned wspace :1; // Word Space flag
unsigned timeout :1; // Timeout flag
unsigned overload :1; // Overload flag
} bflags;
static bool cw_state; // Current decoded signal state
static sigbuf sig[CW_SIG_BUFSIZE]; // A circular buffer of decoded input levels and durations, input from
static int32_t sig_lastrx = 0; // Circular buffer in pointer, updated by SignalSampler
static int32_t sig_incount = 0; // Circular buffer in pointer, copy of sig_lastrx, used by CW Decode functions
static int32_t sig_outcount = 0; // Circular buffer out pointer, used by CW Decode functions
static int32_t sig_timer = 0; // Elapsed time of current signal state, dependent
// on sample rate, decimation factor and CW_DECODE_BLOCK_SIZE
// 48ksps & decimation-by-4 equals 12ksps
// if CW_DECODE_BLOCK_SIZE == 32, then we have 12000/32 = 375 blocks per second, which means
// one Goertzel magnitude is calculated 375 times a second, which means 2.67ms per timer_stepsize
// this is very similar to the original 2.9ms (when using FFT256 in the Teensy 3 original sketch)
// DD4WH 2017_09_08
static int32_t timer_stepsize = 1; // equivalent to 2.67ms, see above
static int32_t cur_time; // copy of sig_timer
static int32_t cur_outcount = 0; // Basically same as sig_outcount, for Error Correction functionality
static int32_t last_outcount = 0; // sig_outcount for previous character, used for Error Correction func
sigbuf data[CW_DATA_BUFSIZE]; // Buffer containing decoded dot/dash and time information
// for assembly into a character
static uint8_t data_len = 0; // Length of incoming character data
static uint32_t code; // Decoded dot/dash info in pairs of bits, - is encoded as 11, and . is encoded as 10
static bflags b; // Various Operational state flags
typedef struct
{
float32_t pulse_avg; // CW timing variables - pulse_avg is a composite value
float32_t dot_avg;
float32_t dash_avg; // Dot and Dash Space averages
float32_t symspace_avg;
float32_t cwspace_avg; // Intra symbol Space and Character-Word Space
int32_t w_space; // Last word space time
} cw_times_t;
static cw_times_t cw_times;
// audio signal buffer
static float32_t raw_signal_buffer[CW_DECODER_BLOCKSIZE_MAX]; //cw_decoder_config.blocksize];
// RINGBUFFER HELPER MACROS START
#define ring_idx_wrap_upper(value,size) (((value) >= (size)) ? (value) - (size) : (value))
#define ring_idx_wrap_zero(value,size) (((value) < (0)) ? (value) + (size) : (value))
/*
* @brief adjust index "value" by "change" while keeping it in the ring buffer size limits of "size"
* @returns the value changed by adding change to it and doing a modulo operation on it for the ring buffer size. So return value is always 0 <= result < size
*/
#define ring_idx_change(value,change,size) (change>0 ? ring_idx_wrap_upper((value)+(change),(size)):ring_idx_wrap_zero((value)+(change),(size)))
#define ring_idx_increment(value,size) ((value+1) == (size)?0:(value+1))
#define ring_idx_decrement(value,size) ((value) == 0?(size)-1:(value)-1)
// Determine number of states waiting to be processed
#define ring_distanceFromTo(from,to) (((to) < (from))? ((CW_SIG_BUFSIZE + (to)) - ((from) )) : (to - from))
// RINGBUFFER HELPER MACROS END
static void CW_Decode_exe(void)
{
bool newstate;
static float32_t CW_env = 0.0;
static float32_t CW_mag = 0.0;
static float32_t CW_noise = 0.0;
float32_t CW_clipped = 0.0;
static float32_t old_siglevel = 0.001;
static float32_t speed_wpm_avg = 0.0;
float32_t siglevel; // signal level from Goertzel calculation
static bool prevstate; // Last recorded state of signal input (mark or space)
// 1.) get samples
// these are already in raw_signal_buffer
// 2.) calculate Goertzel
for (uint16_t index = 0; index < cw_decoder_config.blocksize; index++)
{
AudioFilter_GoertzelInput(&cw_goertzel, raw_signal_buffer[index]);
}
float32_t magnitudeSquared = AudioFilter_GoertzelEnergy(&cw_goertzel);
// I am not sure whether we would need an AGC here, because the audio chain already has an AGC
// Now I am sure, we do not need it
// 3.) AGC
#if 0
float32_t pklvl; // Used for AGC calculations
if (cw_decoder_config.AGC_enable)
{
pklvl = CW_agcvol * CW_vol * magnitudeSquared; // Get level at Goertzel frequency
if (pklvl > AGC_MAX_PEAK)
CW_agcvol = CW_agcvol * CW_AGC_ATTACK; // Decrease volume if above this level.
if (pklvl < AGC_MIN_PEAK)
CW_agcvol = CW_agcvol * CW_AGC_DECAY; // Increase volume if below this level.
if (CW_agcvol > 1.0)
CW_agcvol = 1.0; // Cap max at 1.0
siglevel = CW_agcvol * CW_vol * pklvl;
}
else
#endif
{
siglevel = magnitudeSquared;
}
// 4.) signal averaging/smoothing
#if 0
static float32_t avg_win[20]; // Sliding window buffer for signal averaging, if used
static uint8_t avg_cnt = 0; // Sliding window counter
avg_win[avg_cnt] = siglevel; // Add value onto "sliding window" buffer
avg_cnt = ring_idx_increment(avg_cnt, cw_decoder_config.average);
float32_t lvl = 0; // Multiuse variable
for (uint8_t x = 0; x < cw_decoder_config.average; x++) // Average up all values within sliding window
{
lvl = lvl + avg_win[x];
}
siglevel = lvl / cw_decoder_config.average;
#else
// better use exponential averager for averaging/smoothing here !? Let<65>s try!
// siglevel = siglevel * SIGNAL_TAU + ONEM_SIGNAL_TAU * old_siglevel;
// old_siglevel = magnitudeSquared;
#endif
// 4b.) automatic threshold correction
if(cw_decoder_config.use_3_goertzels)
{
CW_mag = siglevel;
CW_env = decayavg(CW_env, CW_mag, (CW_mag > CW_env)?
// (CW_ONE_BIT_SAMPLE_COUNT / 4) : (CW_ONE_BIT_SAMPLE_COUNT * 16));
(cw_decoder_config.thresh /1000 / 4) : (cw_decoder_config.thresh /1000 * 16));
CW_noise = decayavg(CW_noise, CW_mag, (CW_mag < CW_noise)?
//(CW_ONE_BIT_SAMPLE_COUNT / 4) : (CW_ONE_BIT_SAMPLE_COUNT * 48));
(cw_decoder_config.thresh /1000 / 4) : (cw_decoder_config.thresh /1000 * 48));
CW_clipped = CW_mag > CW_env? CW_env: CW_mag;
if (CW_clipped < CW_noise)
{
CW_clipped = CW_noise;
}
float32_t v1 = (CW_clipped - CW_noise) * (CW_env - CW_noise) -
0.8 * ((CW_env - CW_noise) * (CW_env - CW_noise));
// 0.85 * ((CW_env - CW_noise) * (CW_env - CW_noise));
// ((CW_env - CW_noise) * (CW_env - CW_noise));
// 0.25 * ((CW_env - CW_noise) * (CW_env - CW_noise));
//lowpass
// v1 = RttyDecoder_lowPass(v1, rttyDecoderData.lpfConfig, &rttyDecoderData.lpfData);
siglevel = v1 * SIGNAL_TAU + ONEM_SIGNAL_TAU * old_siglevel;
old_siglevel = v1;
// bool newstate = (siglevel > 0)? false:true;
newstate = (siglevel < 0)? false:true;
}
// 5.) signal state determination
//----------------
// Signal State sampling
// noise cancel requires at least two consecutive samples to be
// of same (changed state) to accept change (i.e. a single sample change is ignored).
else
{
siglevel = siglevel * SIGNAL_TAU + ONEM_SIGNAL_TAU * old_siglevel;
old_siglevel = magnitudeSquared;
newstate = (siglevel >= cw_decoder_config.thresh);
}
if(cw_decoder_config.noisecancel_enable)
{
static bool change; // reads to be the same to confirm a true change
if (change == TRUE)
{
cw_state = newstate;
change = FALSE;
}
else if (newstate != cw_state)
{
change = TRUE;
}
}
else
{// No noise canceling
cw_state = newstate;
}
ads.CW_signal = cw_state;
// if(ts.dmod_mode == DEMOD_CW)
if(cw_decoder_config.show_CW_LED == true && ts.cw_decoder_enable && ts.dmod_mode == DEMOD_CW)
{
Board_RedLed(cw_state == true? LED_STATE_ON : LED_STATE_OFF);
}
// 6.) fill into circular buffer
//----------------
// Record state changes and durations onto circular buffer
if (cw_state != prevstate)
{
// Enter the type and duration of the state change into the circular buffer
sig[sig_lastrx].state = prevstate;
sig[sig_lastrx].time = sig_timer;
// Zero circular buffer when at max
sig_lastrx = ring_idx_increment(sig_lastrx, CW_SIG_BUFSIZE);
sig_timer = 0; // Zero the signal timer.
prevstate = cw_state; // Update state
}
//----------------
// Count signal state timer upwards based on which sampling rate is in effect
sig_timer = sig_timer + timer_stepsize;
if (sig_timer > ONE_SECOND * CW_TIMEOUT)
{
sig_timer = ONE_SECOND * CW_TIMEOUT; // Impose a MAXTIME second boundary for overflow time
}
sig_incount = sig_lastrx; // Current Incount pointer
cur_time = sig_timer;
// 7.) CW Decode
if(ts.cw_decoder_enable && ts.dmod_mode == DEMOD_CW)
{
CW_Decode(); // Do all the heavy lifting
}
// calculation of speed of the received morse signal on basis of the standard "PARIS"
float32_t spdcalc = 10.0 * cw_times.dot_avg + 4.0 * cw_times.dash_avg + 9.0 * cw_times.symspace_avg + 5.0 * cw_times.cwspace_avg;
// update only if initialized and prevent division by zero
if(b.initialized == true && spdcalc > 0)
{
// Convert to Milliseconds per Word
float32_t speed_ms_per_word = spdcalc * 1000.0 / (cw_decoder_config.sampling_freq / (float32_t)cw_decoder_config.blocksize);
float32_t speed_wpm_raw = (0.5 + 60000.0 / speed_ms_per_word); // calculate words per minute
speed_wpm_avg = speed_wpm_raw * 0.3 + 0.7 * speed_wpm_avg; // a little lowpass filtering
}
else
{
speed_wpm_avg = 0; // we have no calculated speed, i.e. not synchronized to signal
}
cw_decoder_config.speed = speed_wpm_avg; // for external use, 0 indicates no signal condition
if(ts.txrx_mode == TRX_MODE_TX)
{ // just to ensure that during RX/TX switching the red LED remains lit in TX_mode
Board_RedLed(LED_STATE_ON);
}
}
void CwDecode_RxProcessor(float32_t * const src, int16_t blockSize)
{
static uint16_t sample_counter = 0;
for (uint16_t idx = 0; idx < blockSize; idx++)
{
raw_signal_buffer[sample_counter] = src[idx];
sample_counter++;
}
if (sample_counter >= cw_decoder_config.blocksize)
{
CW_Decode_exe();
sample_counter = 0;
}
}
//------------------------------------------------------------------
//
// Initialization Function (non-blocking-style)
// Determine Pulse, Dash, Dot and initial
// Character-Word time averages
//
// Input is the circular buffer sig[], including in and out counters
// Output is variables containing dot dash and space averages
//
//------------------------------------------------------------------
static void InitializationFunc(void)
{
static int16_t startpos, progress; // Progress counter, size = SIG_BUFSIZE
static bool initializing = FALSE; // Bool for first time init of progress counter
int16_t processed; // Number of states that have been processed
float32_t t; // We do timing calculations in floating point
// to gain a little bit of precision when low
// sampling rate
// Set up progress counter at beginning of initialize
if (initializing == FALSE)
{
startpos = sig_outcount; // We start at last processed mark/space
progress = sig_outcount;
initializing = TRUE;
cw_times.pulse_avg = 0; // Reset CW timing variables to 0
cw_times.dot_avg = 0;
cw_times.dash_avg = 0;
cw_times.symspace_avg = 0;
cw_times.cwspace_avg = 0;
cw_times.w_space = 0;
}
// Board_RedLed(LED_STATE_ON);
// Determine number of states waiting to be processed
processed = ring_distanceFromTo(startpos,progress);
if (processed >= 98)
{
b.initialized = TRUE; // Indicate we're done and return
initializing = FALSE; // Allow for correct setup of progress if
// InitializaitonFunc is invoked a second time
// Board_RedLed(LED_STATE_OFF);
}
if (progress != sig_incount) // Do we have a new state?
{
t = sig[progress].time;
if (sig[progress].state) // Is it a pulse?
{
if (processed > 32) // More than 32, getting stable
{
if (t > cw_times.pulse_avg)
{
cw_times.dash_avg = cw_times.dash_avg + (t - cw_times.dash_avg) / 4.0; // (e.q. 4.5)
}
else
{
cw_times.dot_avg = cw_times.dot_avg + (t - cw_times.dot_avg) / 4.0; // (e.q. 4.4)
}
}
else // Less than 32, still quite unstable
{
if (t > cw_times.pulse_avg)
{
cw_times.dash_avg = (t + cw_times.dash_avg) / 2.0; // (e.q. 4.2)
}
else
{
cw_times.dot_avg = (t + cw_times.dot_avg) / 2.0; // (e.q. 4.1)
}
}
cw_times.pulse_avg = (cw_times.dot_avg / 4 + cw_times.dash_avg) / 2.0; // Update pulse_avg (e.q. 4.3)
}
else // Not a pulse - determine character_word space avg
{
if (processed > 32)
{
if (t > cw_times.pulse_avg) // Symbol space?
{
cw_times.cwspace_avg = cw_times.cwspace_avg + (t - cw_times.cwspace_avg) / 4.0; // (e.q. 4.8)
}
else
{
cw_times.symspace_avg = cw_times.symspace_avg + (t - cw_times.symspace_avg) / 4.0; // New EQ, to assist calculating Rate
}
}
}
progress = ring_idx_increment(progress,CW_SIG_BUFSIZE); // Increment progress counter
}
}
//------------------------------------------------------------------
//
// Spike Cancel function
//
// Optionally selectable in CWReceive.h, used by Data Recognition
// function to identify and ignore spikes of short duration.
//
//------------------------------------------------------------------
bool CwDecoder_IsSpike(uint32_t t)
{
bool retval = false;
if (cw_decoder_config.spikecancel == CW_SPIKECANCEL_MODE_SPIKE) // SPIKE CANCEL // Squash spikes/transients of short duration
{
retval = t <= CW_SPIKECANCEL_MAX_DURATION;
}
else if (cw_decoder_config.spikecancel == CW_SPIKECANCEL_MODE_SHORT) // SHORT CANCEL // Squash spikes shorter than 1/3rd dot duration
{
retval = (3 * t < cw_times.dot_avg) && (b.initialized == TRUE); // Only do this if we are not initializing dot/dash periods
}
return retval;
}
float32_t spikeCancel(float32_t t)
{
static bool spike;
if (cw_decoder_config.spikecancel != CW_SPIKECANCEL_MODE_OFF)
{
if (CwDecoder_IsSpike(t) == true)
{
spike = TRUE;
sig_outcount = ring_idx_increment(sig_outcount, CW_SIG_BUFSIZE); // If short, then do nothing
t = 0.0;
}
else if (spike == TRUE) // Check if last state was a short Spike or Drop
{
spike = FALSE;
// Add time of last three states together.
t = t
+ sig[ring_idx_change(sig_outcount, -1, CW_SIG_BUFSIZE)].time
+ sig[ring_idx_change(sig_outcount, -2, CW_SIG_BUFSIZE)].time;
}
}
return t;
}
//------------------------------------------------------------------
//
// Data Recognition Function (non-blocking-style)
// Decode dots, dashes and spaces and group together
// into a character.
//
// Input is the circular buffer sig[], including in and out counters
// Variables containing dot, dash and space averages are maintained, and
// output is a data[] buffer containing decoded dot/dash information, a
// data_len variable containing length of incoming character data.
// The function returns TRUE when further calls will not yield a change or a complete new character has been decoded.
// The bool variable the parameter points to is set to true if a new character has been decoded
// In addition, b.wspace flag indicates whether long (word) space after char
//
//------------------------------------------------------------------
bool DataRecognitionFunc(bool* new_char_p)
{
bool not_done = FALSE; // Return value
static bool processed;
*new_char_p = FALSE;
//-----------------------------------
// Do we have a new state to process?
if (sig_outcount != sig_incount)
{
not_done = true;
b.timeout = FALSE; // Mainly used by Error Correction Function
const float32_t t = spikeCancel(sig[sig_outcount].time); // Get time of the new state
// Squash spikes/transients if enabled
// Attention: Side Effect -> sig_outcount has been be incremented inside spikeCancel if result == 0, because of this we increment only if not 0
if (t > 0) // not a spike (or spike processing not enabled)
{
const bool is_markstate = sig[sig_outcount].state;
sig_outcount = ring_idx_increment(sig_outcount, CW_SIG_BUFSIZE); // Update process counter
//-----------------------------------
// Is it a Mark (keydown)?
if (is_markstate == true)
{
processed = FALSE; // Indicate that incoming character is not processed
// Determine if Dot or Dash (e.q. 4.10)
if ((cw_times.pulse_avg - t) >= 0) // It is a Dot
{
b.dash = FALSE; // Clear Dash flag
data[data_len].state = 0; // Store as Dot
cw_times.dot_avg = cw_times.dot_avg + (t - cw_times.dot_avg) / 8.0; // Update cw_times.dot_avg (e.q. 4.6)
}
//-----------------------------------
// Is it a Dash?
else
{
b.dash = TRUE; // Set Dash flag
data[data_len].state = 1; // Store as Dash
if (t <= 5 * cw_times.dash_avg) // Store time if not stuck key
{
cw_times.dash_avg = cw_times.dash_avg + (t - cw_times.dash_avg) / 8.0; // Update dash_avg (e.q. 4.7)
}
}
data[data_len].time = (uint32_t) t; // Store associated time
data_len++; // Increment by one dot/dash
cw_times.pulse_avg = (cw_times.dot_avg / 4 + cw_times.dash_avg) / 2.0; // Update pulse_avg (e.q. 4.3)
}
//-----------------------------------
// Is it a Space?
else
{
bool full_char_detected = true;
if (b.dash == TRUE) // Last character was a dash
{
b.dash = false;
float32_t eq4_12 = t
- (cw_times.pulse_avg
- ((uint32_t) data[data_len - 1].time
- cw_times.pulse_avg) / 4.0); // (e.q. 4.12, corrected)
if (eq4_12 < 0) // Return on symbol space - not a full char yet
{
cw_times.symspace_avg = cw_times.symspace_avg + (t - cw_times.symspace_avg) / 8.0; // New EQ, to assist calculating Rat
full_char_detected = false;
}
else if (t <= 10 * cw_times.dash_avg) // Current space is not a timeout
{
float32_t eq4_14 = t
- (cw_times.cwspace_avg
- ((uint32_t) data[data_len - 1].time
- cw_times.pulse_avg) / 4.0); // (e.q. 4.14)
if (eq4_14 >= 0) // It is a Word space
{
cw_times.w_space = t;
b.wspace = TRUE;
}
}
}
else // Last character was a dot
{
// (e.q. 4.11)
if ((t - cw_times.pulse_avg) < 0) // Return on symbol space - not a full char yet
{
cw_times.symspace_avg = cw_times.symspace_avg + (t - cw_times.symspace_avg) / 8.0; // New EQ, to assist calculating Rate
full_char_detected = false;
}
else if (t <= 10 * cw_times.dash_avg) // Current space is not a timeout
{
cw_times.cwspace_avg = cw_times.cwspace_avg + (t - cw_times.cwspace_avg) / 8.0; // (e.q. 4.9)
// (e.q. 4.13)
if ((t - cw_times.cwspace_avg) >= 0) // It is a Word space
{
cw_times.w_space = t;
b.wspace = TRUE;
}
}
}
// Process the character
if (full_char_detected == true && processed == FALSE)
{
*new_char_p = TRUE; // Indicate there is a new char to be processed
}
}
}
}
//-----------------------------------
// Long key down or key up
else if (cur_time > (10 * cw_times.dash_avg))
{
// If current state is Key up and Long key up then Char finalized
if (sig[sig_incount].state == false && processed == false)
{
processed = TRUE;
b.wspace = TRUE;
b.timeout = TRUE;
*new_char_p = TRUE; // Process the character
}
}
if (data_len > CW_DATA_BUFSIZE - 2)
{
data_len = CW_DATA_BUFSIZE - 2; // We're receiving garble, throw away
}
if (*new_char_p) // Update circular buffer pointers for Error function
{
last_outcount = cur_outcount;
cur_outcount = sig_outcount;
}
return not_done; // FALSE if all data processed or new character, else TRUE
}
//------------------------------------------------------------------
//
// The Code Generation Function converts the received
// character to a string code[] of dots and dashes
//
//------------------------------------------------------------------
void CodeGenFunc()
{
uint8_t a;
code = 0;
for (a = 0; a < data_len; a++)
{
code *= 4;
if (data[a].state)
{
code += 3; // Dash
}
else
{
code += 2; // Dit
}
}
data_len = 0; // And make ready for a new Char
}
void lcdLineScrollPrint(char c)
{
UiDriver_TextMsgPutChar(c);
}
//------------------------------------------------------------------
//
// The Print Character Function prints to LCD and Serial (USB)
//
//------------------------------------------------------------------
void PrintCharFunc(uint8_t c)
{
//--------------------------------------
//--------------------------------------
// Print Characters to LCD
//--------------------------------------
// Prosigns
if (c == '}')
{
lcdLineScrollPrint('c');
lcdLineScrollPrint('t');
}
else if (c == '(')
{
lcdLineScrollPrint('k');
lcdLineScrollPrint('n');
}
else if (c == '&')
{
lcdLineScrollPrint('a');
lcdLineScrollPrint('s');
}
else if (c == '~')
{
lcdLineScrollPrint('s');
lcdLineScrollPrint('n');
}
else if (c == '>')
{
lcdLineScrollPrint('s');
lcdLineScrollPrint('k');
}
else if (c == '+')
{
lcdLineScrollPrint('a');
lcdLineScrollPrint('r');
}
else if (c == '^')
{
lcdLineScrollPrint('b');
lcdLineScrollPrint('k');
}
else if (c == '{')
{
lcdLineScrollPrint('c');
lcdLineScrollPrint('l');
}
else if (c == '^')
{
lcdLineScrollPrint('a');
lcdLineScrollPrint('a');
}
else if (c == '%')
{
lcdLineScrollPrint('n');
lcdLineScrollPrint('j');
}
else if (c == 0x7f)
{
lcdLineScrollPrint('e');
lcdLineScrollPrint('r');
lcdLineScrollPrint('r');
}
//--------------------------------------
// # is our designated ERROR Symbol
else if (c == 0xff)
{
lcdLineScrollPrint('#');
}
//--------------------------------------
// Normal Characters
/* if (c == 0xfe || c == 0xff)
{
lcdLineScrollPrint('#');
}
*/
else
{
lcdLineScrollPrint(c);
}
}
//------------------------------------------------------------------
//
// The Word Space Function takes care of Word Spaces
// to LCD and Serial (USB).
// Word Space Correction is applied if certain characters, which
// are less likely to be at the end of a word, are received
// The characters tested are applicable to the English language
//
//------------------------------------------------------------------
void WordSpaceFunc(uint8_t c)
{
if (b.wspace == TRUE) // Print word space
{
b.wspace = FALSE;
// Word space correction routine - longer space required if certain characters
if ((c == 'I') || (c == 'J') || (c == 'Q') || (c == 'U') || (c == 'V')
|| (c == 'Z'))
{
int16_t x = (cw_times.cwspace_avg + cw_times.pulse_avg) - cw_times.w_space; // (e.q. 4.15)
if (x < 0)
{
lcdLineScrollPrint(' ');
}
}
else
{
lcdLineScrollPrint(' ');
}
}
}
//------------------------------------------------------------------
//
// Error Correction Function has three parts
// 1) Exits with Error if character is too long (DATA_BUFSIZE-2)
// 2) If a dot duration is determined to be less than half expected,
// then this dot is eliminated by adding it and the two spaces on
// either side to for a new space duration, then new code is generated
// for pattern parsing.
// 3) If not 2) then separate two run-on characters caused by
// a short character space - Extend the char space and reprocess
//
// If not able to resolve anything, then return FALSE
// Return TRUE if something was resolved.
//
//------------------------------------------------------------------
bool ErrorCorrectionFunc(void)
{
bool result = FALSE; // Result of Error resolution - FALSE if nothing resolved
if (data_len >= CW_DATA_BUFSIZE - 2) // Too long char received
{
PrintCharFunc(0xff); // Print Error to LCD and Serial (USB)
WordSpaceFunc(0xff); // Print Word Space to LCD and Serial when required
}
else
{
b.wspace = FALSE;
//-----------------------------------------------------
// Find the location of pulse with shortest duration
// and the location of symbol space of longest duration
int32_t temp_outcount = last_outcount; // Grab a copy of endpos for last successful decode
int32_t slocation = last_outcount; // Long symbol space duration and location
int32_t plocation = last_outcount; // Short pulse duration and location
uint32_t pduration = UINT32_MAX; // Very high number to decrement for min pulse duration
uint32_t sduration = 0; // and a zero to increment for max symbol space duration
// if cur_outcount is < CW_SIG_BUFSIZE, loop must terminate after CW_SIG_BUFSIZE -1 steps
while (temp_outcount != cur_outcount)
{
//-----------------------------------------------------
// Find shortest pulse duration. Only test key-down states
if (sig[temp_outcount].state)
{
bool is_shortest_pulse = sig[temp_outcount].time < pduration;
// basic test -> shorter than all previously seen ones
bool is_not_spike = CwDecoder_IsSpike(sig[temp_outcount].time) == false;
if (is_shortest_pulse == true && is_not_spike == true)
{
pduration = sig[temp_outcount].time;
plocation = temp_outcount;
}
}
//-----------------------------------------------------
// Find longest symbol space duration. Do not test first state
// or last state and only test key-up states
if ((temp_outcount != last_outcount)
&& (temp_outcount != (cur_outcount - 1))
&& (!sig[temp_outcount].state))
{
if (sig[temp_outcount].time > sduration)
{
sduration = sig[temp_outcount].time;
slocation = temp_outcount;
}
}
temp_outcount = ring_idx_increment(temp_outcount,CW_SIG_BUFSIZE);
}
uint8_t decoded[] = { 0xff, 0xff };
//-----------------------------------------------------
// Take corrective action by dropping shortest pulse
// if shorter than half of cw_times.dot_avg
// This can result in one or more valid characters - or Error
if ((pduration < cw_times.dot_avg / 2) && (plocation != temp_outcount))
{
// Add up duration of short pulse and the two spaces on either side,
// as space at pulse location + 1
sig[ring_idx_change(plocation, +1, CW_SIG_BUFSIZE)].time =
sig[ring_idx_change(plocation, -1, CW_SIG_BUFSIZE)].time
+ sig[plocation].time
+ sig[ring_idx_change(plocation, +1, CW_SIG_BUFSIZE)].time;
// Shift the preceding data forward accordingly
temp_outcount = ring_idx_change(plocation, -2 ,CW_SIG_BUFSIZE);
// if last_outcount is < CW_SIG_BUFSIZE, loop must terminate after CW_SIG_BUFSIZE -1 steps
while (temp_outcount != last_outcount)
{
sig[ring_idx_change(temp_outcount, +2, CW_SIG_BUFSIZE)].time =
sig[temp_outcount].time;
sig[ring_idx_change(temp_outcount, +2, CW_SIG_BUFSIZE)].state =
sig[temp_outcount].state;
temp_outcount = ring_idx_decrement(temp_outcount,CW_SIG_BUFSIZE);
}
// And finally shift the startup pointer similarly
sig_outcount = ring_idx_change(last_outcount, +2,CW_SIG_BUFSIZE);
//
// Now we reprocess
//
// Pull out a character, using the adjusted sig[] buffer
// Process character delimited by character or word space
bool dummy;
while (DataRecognitionFunc(&dummy))
{
// nothing
}
CodeGenFunc(); // Generate a dot/dash pattern string
decoded[0] = CwGen_CharacterIdFunc(code); // Convert dot/dash data into a character
if (decoded[0] != 0xff)
{
PrintCharFunc(decoded[0]);
result = TRUE; // Error correction had success.
}
else
{
PrintCharFunc(0xff);
}
}
//-----------------------------------------------------
// Take corrective action by converting the longest symbol space to character space
// This will result in two valid characters - or Error
else
{
// Split char in two by adjusting time of longest sym space to a char space
sig[slocation].time =
((cw_times.cwspace_avg - 1) >= 1 ? cw_times.cwspace_avg - 1 : 1); // Make sure it is always larger than 0
sig_outcount = last_outcount; // Set circ buffer reference to the start of previous failed decode
//
// Now we reprocess
//
// Debug - If timing is out of whack because of noise, with rate
// showing at >99 WPM, then DataRecognitionFunc() occasionally fails.
// Not found out why, but millis() is used to guards against it.
// Process first character delimited by character or word space
bool dummy;
while (DataRecognitionFunc(&dummy))
{
// nothing
}
CodeGenFunc(); // Generate a dot/dash pattern string
decoded[0] = CwGen_CharacterIdFunc(code); // Convert dot/dash pattern into a character
// Process second character delimited by character or word space
while (DataRecognitionFunc(&dummy))
{
// nothing
}
CodeGenFunc(); // Generate a dot/dash pattern string
decoded[1] = CwGen_CharacterIdFunc(code); // Convert dot/dash pattern into a character
if ((decoded[0] != 0xff) && (decoded[1] != 0xff)) // If successful error resolution
{
PrintCharFunc(decoded[0]);
PrintCharFunc(decoded[1]);
result = TRUE; // Error correction had success.
}
else
{
PrintCharFunc(0xff);
}
}
}
return result;
}
//------------------------------------------------------------------
//
// CW Decode manages all the decode Functions.
// It establishes dot/dash/space periods through the Initialization
// function, and when initialized (or if excessive time when not fully
// initialized), then it runs DataRecognition, CodeGen and CharacterId
// functions to decode any incoming data. If not successful decode
// then ErrorCorrection is attempted, and if that fails, then
// Initialization is re-performed.
//
//------------------------------------------------------------------
void CW_Decode(void)
{
//-----------------------------------
// Initialize pulse_avg, dot_avg, cw_times.dash_avg, cw_times.symspace_avg, cwspace_avg
if (b.initialized == FALSE)
{
InitializationFunc();
}
//-----------------------------------
// Process the works once initialized - or if timeout
if ((b.initialized == TRUE) || (cur_time >= ONE_SECOND * CW_TIMEOUT)) //
{
bool received; // True on a symbol received
DataRecognitionFunc(&received); // True if new character received
if (received && (data_len > 0)) // also make sure it is not a spike
{
CodeGenFunc(); // Generate a dot/dash pattern string
uint8_t decoded = CwGen_CharacterIdFunc(code);
// Identify the Character
// 0xff if char not recognized
if (decoded < 0xfe) // 0xfe = spike suppression, 0xff = error
{
PrintCharFunc(decoded); // Print to LCD and Serial (USB)
WordSpaceFunc(decoded); // Print Word Space to LCD and Serial when required
}
else if (decoded == 0xff) // Attempt Error Correction
{
// If Error Correction function cannot resolve, then reinitialize speed
if (ErrorCorrectionFunc() == FALSE)
{
b.initialized = FALSE;
}
}
}
}
}
void CwDecoder_WpmDisplayClearOrPrepare(bool prepare)
{
uint16_t color1 = prepare?White:Black;
uint16_t color2 = prepare?Green:Black;
UiLcdHy28_PrintText(ts.Layout->CW_DECODER_WPM.x, ts.Layout->CW_DECODER_WPM.y," --",color1,Black,0);
UiLcdHy28_PrintText(ts.Layout->CW_DECODER_WPM.x + 27, ts.Layout->CW_DECODER_WPM.y, "wpm", color2, Black, 4);
if (prepare == true)
{
CwDecoder_WpmDisplayUpdate(true);
}
}
void CwDecoder_WpmDisplayUpdate(bool force_update)
{
static uint8_t old_speed = 0;
if(cw_decoder_config.speed != old_speed || force_update == true)
{
char WPM_str[10];
snprintf(WPM_str, 10, cw_decoder_config.speed > 0? "%3u" : " --", cw_decoder_config.speed);
UiLcdHy28_PrintText(ts.Layout->CW_DECODER_WPM.x, ts.Layout->CW_DECODER_WPM.y, WPM_str,White,Black,0);
}
}