// Prerelease version, 8/23/05
// Author: Jim Battle

// This program reads a .WAV file and attempts to decode the data stream
// off that recording.  The audio format must be in RIFF WAV format,
// non-compressed, one or two channels.  It allows for mono/stereo input,
// any input sampling rate (lower sampling rates increase the decoding
// error rate), and 8b or 16b samples.

// The Datapoint 2200 records data bits using a FSK method, namely one
// half cycle at frequency X means a one bit and one full cycle at
// frequency 2X means a zero bit.  This is stated this way because
// the samples processed here are off a tape recorder running at 1 7/8 ips
// while the datapoint runs its tape at 7.5 ips, a factor of four times
// faster.  To be specific, at this reduced speed, a one bit is a half cycle
// of 481.25 Hz and a zero bit is a full cycle of 962.5 Hz.  The original
// 7.5 ips frequencies are 1925 Hz and 3850 Hz.  The preamble while the
// tape is coming up to speed is a train of 1 bits.
//
// The structure of the data on a tape is as follows:
//
//    (1) bytes are eight bits
//
//    (2) a sync pattern of 010 brackets each byte, so for example
//        two bytes would be (sync)(byte0)(sync)(byte1)(sync).  thus
//        the character rate is 350 cps at 7.5 ips.
//
//    (3) in the forward direction, bits of a byte are recorded
//        msb first, lsb last.
//
//    (4) a record is a sequence of bytes with valid sync patterns
//        followed by an all 1's byte and an invalid sync (of 111)
//
//    (5) the inter-record gap time is about 280 ms, or 98 character times
//
//    (6) trains of one bits are recorded before and after a record to
//        allow a PLL to track the signal and to form the end of record
//        marker at both sides of a byte of 1 bits and a sync of 111.

// There isn't a whole lot of high-powered thinking going on here.
// It is just a lot of guess work on my part as to what seems to work
// and what doesn't.  Start simple and add complexity as required.
//
// The program is structured to operate on a rolling window through
// the file, rather than doing the easier thing of sucking in the
// whole file and then operating on it.  This is because the files
// can last a few minutes, which would consume a prohibitive amount
// of memory.

#include <stdio.h>
#include <stdlib.h>
#include <stdarg.h>	// for varargs
#include <string.h>
#include <assert.h>

// define as "1" on little endian architectures, otherwise define as "0"
// FIXME: do this automatically
#define LITTLE_ENDIAN 1

// support dumping records as intel hex files
#define SUPPORT_HEX 1

// at least this many "1" bits must be s seen before a sync pattern in order
// to be taken seriously.  this is required because as the tape speed slows
// at the end of a record, sometimes false syncs are detected.
#define SYNC_THRESHOLD  32

// ========================================================================
// type definitions

typedef unsigned long  uint32;
typedef          long  int32;
typedef unsigned short uint16;
typedef          short int16;
typedef unsigned char  uint8;
typedef          char  int8;

typedef int16 sample_t;

#define MIN(a,b) (((a)<(b))?(a):(b))
#define MAX(a,b) (((a)>(b))?(a):(b))
#define ABS(a)   (((a)<(0))?(-a):(a))

// ========================================================================
// global variables

// command line options
int   opt_v;		// verbosity level
bool  opt_x;		// explode mode
int   opt_x_num;	// chunk number
char *opt_ofn;		// output filename
char *opt_ifn;		// input filename
enum { OFMT_TAP=0, OFMT_BIN, OFMT_HEX } opt_ofmt;

// .wav file attributes
bool   inmono;			// input  file is mono (1) or stereo (0)
int    sample_bytes;		// number of bytes per sample
uint32 expected_samples;	// number of samples in file
uint32 sample_rate;		// samples/second

FILE  *fIn;		// input audio file handle

// period of a zero bit, in Hz, at 1 7/8 ips
const float zero_freq = 962.5f;

// pll lock range = nominal +/- 25%
const float lock_range = 0.25f;
float min_samples_per_bit;
float max_samples_per_bit;

float samples_per_bit;		// bit period, in samples
float pll_period;		// current PLL estimate of bit period


// ========================================================================
// filtered reporting

void
tfprintf(int verbosity, FILE *fp, char *fmt, ...)
{
    char buff[1000];
    va_list args;

    if (verbosity <= opt_v) {
	va_start(args, fmt);
	_vsnprintf(buff, sizeof(buff), fmt, args);
	va_end(args);
	fputs(buff, fp);
    }
}

void
tprintf(int verbosity, char *fmt, ...)
{
    char buff[1000];
    va_list args;

    if (verbosity <= opt_v) {
	va_start(args, fmt);
	_vsnprintf(buff, sizeof(buff), fmt, args);
	va_end(args);
	fputs(buff, stdout);
    }
}

// ========================================================================
// WAV file parsing

// RIFF WAV file format
//  __________________________
// | RIFF WAVE Chunk          |
// |   groupID  = 'RIFF'      |
// |   riffType = 'WAVE'      |
// |    __________________    |
// |   | Format Chunk     |   |
// |   |   ckID = 'fmt '  |   |
// |   |__________________|   |
// |    __________________    |
// |   | Sound Data Chunk |   |
// |   |   ckID = 'data'  |   |
// |   |__________________|   |
// |__________________________|
//
// although it is legal to have more than one data chunk, this
// program assumes there is only one.

const uint32 RiffID = ('R' <<  0) | ('I' << 8) | ('F' << 16) | ('F' << 24);
const uint32 WaveID = ('W' <<  0) | ('A' << 8) | ('V' << 16) | ('E' << 24);
typedef struct {
    uint32	groupID;	// should be 'RIFF'
    uint32	riffBytes;	// number of bytes in file after this header
    uint32	riffType;	// should be 'WAVE'
} RIFF_t;

const uint32 FmtID = ('f' <<  0) | ('m' << 8) | ('t' << 16) | (' ' << 24);
typedef struct {
    uint32	chunkID;		// should be 'fmt '
    int32	chunkSize;		// not including first 8 bytes of header
    int16	FormatTag;		// 1=uncompressed
    uint16	Channels;		// number of audio channels
    uint32	Frequency;		// sample frequency
    uint32	AvgBPS;			// we'll ignore this
    uint16	BlockAlign;		// we'll ignore this
    uint16	BitsPerSample;
} FormatChunk_t;

const uint32 DataID = ('d' <<  0) | ('a' << 8) | ('t' << 16) | ('a' << 24);
typedef struct {
    uint32	chunkID;	// must be 'data'
    int32	chunkSize;	// not including first 8 bytes of header
//  unsigned char data[];	// everything that follows
} DataChunk_t;


#define SWAP16(v) ( ((((uint16)(v)) & 0x00FF) << 8) | \
		    ((((uint16)(v)) & 0xFFFF) >> 8) )

#define SWAP32(v) ( ((((uint32)(v)) & 0x000000FF) << 24) | \
		    ((((uint32)(v)) & 0x0000FF00) <<  8) | \
		    ((((uint32)(v)) & 0x00FF0000) >>  8) | \
		    ((((uint32)(v)) & 0xFF000000) >> 24) )

// make sure the file just opened is a WAV file, and if so,
// read some of the critical parameters
void
CheckHeader(void)
{
    RIFF_t        RiffHdr;
    FormatChunk_t FormatHdr;
    DataChunk_t   DataHdr;

    if (fread(&RiffHdr, sizeof(RiffHdr), 1, fIn) != 1) {
	fprintf(stderr, "Error: file didn't contain a RIFF header\n");
	exit(-1);
    }

#if !LITTLE_ENDIAN
    // do endian swap
    RiffHdr.groupID   = (uint32)SWAP32(RiffHdr.groupID);
    RiffHdr.riffBytes = (uint32)SWAP32(RiffHdr.riffBytes);
    RiffHdr.riffType  = (uint32)SWAP32(RiffHdr.riffType);
#endif

    if ((RiffHdr.groupID  != RiffID) ||
	(RiffHdr.riffType != WaveID)) {
	fprintf(stderr, "Error: input file not a WAV file\n");
	exit(-1);
    }

    if (fread(&FormatHdr, sizeof(FormatHdr), 1, fIn) != 1) {
	fprintf(stderr, "Error: file didn't contain a format block\n");
	exit(-1);
    }

#if !LITTLE_ENDIAN
    // do endian swap
    FormatHdr.chunkID       = (uint32)SWAP32(FormatHdr.chunkID);
    FormatHdr.chunkSize     = ( int32)SWAP32(FormatHdr.chunkSize);
    FormatHdr.FormatTag     = ( int16)SWAP16(FormatHdr.FormatTag);
    FormatHdr.Channels      = (uint16)SWAP16(FormatHdr.Channels);
    FormatHdr.Frequency     = (uint32)SWAP32(FormatHdr.Frequency);
    FormatHdr.AvgBPS        = (uint32)SWAP32(FormatHdr.AvgBPS);
    FormatHdr.BlockAlign    = (uint16)SWAP16(FormatHdr.BlockAlign);
    FormatHdr.BitsPerSample = (uint16)SWAP16(FormatHdr.BitsPerSample);
#endif

    if ((FormatHdr.chunkID != FmtID) ||
        (FormatHdr.chunkSize != sizeof(FormatHdr)-8)) {
	fprintf(stderr, "Error: I can't deal with this type of WAV file\n");
	exit(-1);
    }

    if (FormatHdr.BitsPerSample == 8)
	sample_bytes = 1;
    else if (FormatHdr.BitsPerSample == 16)
	sample_bytes = 2;
    else {
	fprintf(stderr, "Error: samples must be either 8b or 16b\n");
	exit(-1);
    }

    if (FormatHdr.FormatTag != 1) {
	fprintf(stderr, "Error: can't deal with compressed data\n");
	exit(-1);
    }

    if (FormatHdr.Channels != 1 && FormatHdr.Channels != 2) {
	fprintf(stderr, "Error: can't handle too many channels\n");
	exit(-1);
    }
    inmono = (FormatHdr.Channels == 1);

    sample_rate = FormatHdr.Frequency;
    if (FormatHdr.Frequency < 11000) {
	fprintf(stderr, "Warning: the sample rate is low -- it might hurt conversion\n");
	exit(-1);
    }

    if (fread(&DataHdr, sizeof(DataHdr), 1, fIn) != 1) {
	fprintf(stderr, "Error: file didn't contain a DATA header\n");
	exit(-1);
    }

    if (DataHdr.chunkID != DataID) {
	fprintf(stderr, "Error: I can't deal with this type of WAV file (2)\n");
	exit(-1);
    }

    // compute dependent parameters
    expected_samples = DataHdr.chunkSize / 
		       (sample_bytes * FormatHdr.Channels);

    // period, in samples
    samples_per_bit = (float)sample_rate / zero_freq;

    // pll lock range = nominal +/- 25%
    pll_period = samples_per_bit;	// start at nominal
    min_samples_per_bit = samples_per_bit * (1.0f - lock_range);
    max_samples_per_bit = samples_per_bit * (1.0f + lock_range);

    float sec = (float)expected_samples/sample_rate;
    tprintf(1, "File: '%s'\n", opt_ifn);
    tprintf(1, "WAV format: %d samples/sec, %db %s\n", sample_rate,
	       8*sample_bytes, inmono ? "mono" : "stereo");
    tprintf(1, "Expected # of samples: %ld", expected_samples);
    tprintf(1, " (%.2f seconds)\n",sec);
}


// return a mono sample from the input file.
// if the input file is in stereo, it averages the two channels.
// this routine reads blocks for efficiency, but doles out one sample
// per request.
sample_t
GetMonoSample(void)
{
    sample_t newsamp;

    if (inmono && (sample_bytes == 1)) {
	// mono 8b
	int8 b0 = (int8)fgetc(fIn) - 128;
	newsamp = (sample_t)(b0 << 8);
    } else if (inmono && (sample_bytes == 2)) {
	uint8 b0 = (uint8)fgetc(fIn);
	int8  b1 =  (int8)fgetc(fIn);
	newsamp = (sample_t)((b1<<8) + b0);
    } else if (!inmono && (sample_bytes == 1)) {
	// 8b stereo
	int8 b0 = fgetc(fIn) - 128;
	int8 b1 = fgetc(fIn) - 128;
	newsamp = (sample_t)((b0+b1) << 7);
    } else if (!inmono && (sample_bytes == 2)) {
	// 16b stereo
	uint8 b0 = (uint8)fgetc(fIn);
	int8  b1 =  (int8)fgetc(fIn);
	uint8 b2 = (uint8)fgetc(fIn);
	int8  b3 =  (int8)fgetc(fIn);
	sample_t left  = (sample_t)((b1<<8) + b0);
	sample_t right = (sample_t)((b3<<8) + b2);
	newsamp = (left+right+1)>>1;
    }

    return newsamp;
}

// =========================================================================
// we maintain in memory a WORKBUFSIZE window of samples.
// the tricky part is that the start of the buffer might be anywhere
// in the buffer, and all other addressing is modulo the buffer size.
//
// if any access is made to the last BUMP samples, we roll the window
// over by QTRWORKBUF samples.

// number of samples we are holding in memory
#define WORKBUFSIZE 16384		// make this a power of two
#define WORKBUFMASK (WORKBUFSIZE-1)	// for modulo wrapping
#define HALFWORKBUF (WORKBUFSIZE/2)
#define QTRWORKBUF  (WORKBUFSIZE/4)
#define BUMP 1024

int32 windowstart;	// the oldest sample in the buffer
int32 windowoffset;	// where in the window the oldest sample lives

// this holds the samples we are working on
sample_t inbuf[WORKBUFSIZE];


int
sample_addr(int32 n)
{
#if 1
    if (n < windowstart || n >= windowstart+WORKBUFSIZE) {
	fprintf(stderr, "Error: sample %d, workbuf[] access out of range\n", n);
	exit(-1);
    }
#endif

    // see if we are bumping into the end of the buffer
    if (n >= windowstart + WORKBUFSIZE - BUMP) {
	// move the input buffer window forward 'n' samples.
	// we do this through modulo address arithmetic;
	// we don't actually move the samples around.

	// adjust pointers
	windowstart  += QTRWORKBUF;
	windowoffset = (QTRWORKBUF + windowoffset) & WORKBUFMASK;

	// fill up newly exposed portion of buffer
	int off = (windowoffset + 3*QTRWORKBUF) & WORKBUFMASK;
	for(int32 t=off; t<off+QTRWORKBUF; t++)
	    inbuf[t] = GetMonoSample();
    }

    return (n - windowstart + windowoffset) & WORKBUFMASK;
}

#define GETIN(n) (inbuf[sample_addr(n)])


// at time zero, we initialize the first half of the buffer
// to whatever value the first sample has, then fill the rest
// of the buffer from samples from the file.
void
FillInbuffer(void)
{
    sample_t s;
    int t;

    // at time zero, we put the first sample right in the middle
    windowstart  = -HALFWORKBUF;
    windowoffset = 0;

    // fill the first half of the buffer with the first file sample
    s = GetMonoSample();
    for(t=0; t<=HALFWORKBUF; t++)
	inbuf[t] = s;

    // now fill up the rest
    for(t=HALFWORKBUF+1; t<WORKBUFSIZE; t++)
	inbuf[t] = GetMonoSample();
}


// =========================================================================
// file output routines
//
// these routines accept notification of
//    (1) start of a block
//    (2) next received byte
//    (3) error
//    (4) end of block
// based on those events and the global option settings, they
// emit one or more files in the specified format, which may be
//    (a) .tap format
//    (b) .bin format
#if SUPPORT_HEX
//    (c) .hex format
#endif

// i'm not sure what the maximum sized block might be.
// from a small sample of tapes, the empirical maximum is 512 bytes.
// this is much larger than that of course, and is the same size as
// the maximum size memory in a 2200.
#define OBUFFER_SIZE 16384
uint8 obuff_buffer[OBUFFER_SIZE];
int obuff_bytecnt;	// put pointer into obuff_buffer[]
FILE *obuff_fp = NULL;

void
StreamStart(void)
{
    obuff_bytecnt = 0;
}


void
StreamByte(uint8 byte)
{
    assert(obuff_bytecnt < OBUFFER_SIZE);
    obuff_buffer[obuff_bytecnt++] = byte;
}


void
StreamWriteBlkLen(uint32 v)
{
    fputc( ((v >>  0) & 0xFF), obuff_fp );
    fputc( ((v >>  8) & 0xFF), obuff_fp );
    fputc( ((v >> 16) & 0xFF), obuff_fp );
    fputc( ((v >> 24) & 0xFF), obuff_fp );
}

void
StreamEnd(void)
{
    int len = strlen(opt_ofn) + 3;	// +3 for safety
    char *ofn = (char *)malloc(len);
    assert(ofn != NULL);
    int n;

    // when to open the file
    bool open_it =  opt_x ||			// explode requested
		    (opt_ofmt != OFMT_TAP) ||	// bin or hex
		    (opt_x_num == 0);		// first

    // when to close the file
    bool close_it = opt_x ||			// explode requested
		    (opt_ofmt != OFMT_TAP);	// bin or hex

    if (opt_x) {
	sprintf(ofn, opt_ofn, opt_x_num);	// serialize filename
    } else {
	strcpy(ofn, opt_ofn);
    }
    opt_x_num++;

    if (open_it) {
	// open a file
	obuff_fp = fopen(ofn, "wb");
	if (obuff_fp == NULL) {
	    fprintf(stderr, "Error: couldn't open file '%s'\n", ofn);
	    exit(-1);
	}
    }

    switch (opt_ofmt) {

	case OFMT_BIN:	// binary
	    for(n=0; n<obuff_bytecnt; n++) {
		int r = fputc(obuff_buffer[n], obuff_fp);
		assert(r != EOF);
	    }
	    break;

	case OFMT_TAP:	// simh format
	    // from somewhere on the web:
	    //   In this format each tape record is preceded and followed by
	    //   a 4-byte count of the number of bytes in the record. Thus,
	    //   between two records there will be the count for the previous
	    //   record and the count for the next record.
	    // another source confirms this, but with the caveat that zero
	    // length blocks don't repeat the length twice.
	    StreamWriteBlkLen(obuff_bytecnt);
	    for(n=0; n<obuff_bytecnt; n++) {
		int r = fputc(obuff_buffer[n], obuff_fp);
		assert(r != EOF);
	    }
	    if (obuff_bytecnt != 0)
		StreamWriteBlkLen(obuff_bytecnt);
	    break;

#if SUPPORT_HEX
	case OFMT_HEX:	// intel hex format
	    for(n=0; n<obuff_bytecnt; n += 16) {
		int bytesleft = MIN(16, obuff_bytecnt - n);
		int cksum;
		fprintf(obuff_fp, ":%02X%04X%02X", bytesleft, n, 0x00);
		cksum = bytesleft
		      + (n >>  0) + (n >>  8) + (n >> 16) + (n >> 24)
		      + 0x00;
		for(int nn=0; nn<bytesleft; nn++) {
		    uint8 b = obuff_buffer[n+nn];
		    fprintf(obuff_fp, "%02X", b);
		    cksum += b;
		}
		fprintf(obuff_fp, "%02X\n", (256 - cksum) & 0xFF);
	    }
	    fprintf(obuff_fp, ":00000001FF\n");
	    break;
#endif

	default:
	    assert(0);
    }

    if (close_it) {
	fclose(obuff_fp);
	obuff_fp = NULL;
    }

    free(ofn);
}


void
StreamError(int code)
{
    // we sometimes get erroneous syncs as the tape speed slows down if we
    // don't see at least one valid byte after the sync, don't save the block
    if (obuff_bytecnt > 0)
	StreamEnd();
}


void
StreamDone(void)
{
    if (obuff_fp != NULL)
	fclose(obuff_fp);
}

// =========================================================================
// bitstream decoder

#define BH_SIZE (128)		// size of bithistory buffer
#define BH_MASK (BH_SIZE-1)

enum {  BS_LOST=0,
	BS_PREAMBLE,		// in a train of 1 bits
	BS_PREAMBLE_0,		// train of 1s followed by 0
	BS_PREAMBLE_01,		// train of 1s followed by 0,1
	BS_BYTE,		// decoding byte stream
	BS_GAP };

int BSstate = BS_LOST;
int BSbyteCount;

void
Bit(uint32 time, int bit)
{
    static int count;
    static int bits;

    static int bithistory[BH_SIZE];
    static int bh_put = 0;
    static int bh_get = 0;
    static int bh_vld = 0;

    tprintf(3, "sample %d: decoded bit %d\n", time, bit);

    bithistory[bh_put] = bit;
    bh_put = (bh_put + 1) & BH_MASK;
    bh_vld = MIN(bh_vld+1, BH_SIZE);

    switch (BSstate) {

	case BS_LOST:
	    if (bit == 1) {
		BSstate = BS_PREAMBLE;
		count = 1;
	    }
	    break;

	case BS_PREAMBLE:
	    if (bit == 1) {
		count++;
	    } else if ((bit == 0) && (count > SYNC_THRESHOLD))
		BSstate = BS_PREAMBLE_0;
	    else
		BSstate = BS_LOST;
	    break;

	case BS_PREAMBLE_0:
	    if (bit == 1) {
		BSstate = BS_PREAMBLE_01;
	    } else {
		tprintf(2, "sample %d: preamble lost after %d bits\n", time, count+2);
		BSstate = BS_LOST;
	    }
	    break;

	case BS_PREAMBLE_01:
	    if (bit == 0) {
		tprintf(1, "sample %d: preamble and sync after %d bits\n", time, count+3);
		BSstate = BS_BYTE;
		count = 0;
		BSbyteCount = 0;
		bits  = 0x00;
		StreamStart();
	    } else {
		tprintf(2, "sample %d: preamble lost after %d bit %d\n", time, count+3);
		BSstate = BS_PREAMBLE;
		count = 2;
	    }
	    break;

	case BS_BYTE:
	    bits = (bits << 1) | (bit > 0);
	    count++;
	    if (count == 11) {
		bits &= 0x7FF;
		if ((bits & 7) == 2) {  // ... 010 sync code
		    bits = (bits >> 3) & 0xFF;
		    // in fwd direction, data comes off the tape lsb first
		    bits = ((bits & 0x01) << 7)
			 | ((bits & 0x02) << 5)
			 | ((bits & 0x04) << 3)
			 | ((bits & 0x08) << 1)
			 | ((bits & 0x10) >> 1)
			 | ((bits & 0x20) >> 3)
			 | ((bits & 0x40) >> 5)
			 | ((bits & 0x80) >> 7);
		    tprintf(2, "sample %d: byte 0x%02X (%03o)\n", time, bits, bits);
		    StreamByte(bits);
		    count = 0;
		    BSbyteCount++;
		} else if (bits == 0x7FF) {
		    tprintf(1, "sample %d: hit valid gap after %d bytes, PLL period=%f\n", time, BSbyteCount, pll_period);
		    StreamEnd();
		    count = 0;
		    BSstate = BS_GAP;
		} else {
		    if (BSbyteCount == 0) {
			tprintf(1, "sample %d: bad sync code %03X; apparently it was not a valid sync\n", time, bits & 7);
		    } else {
			tprintf(1, "sample %d: bad sync code %03X\n", time, bits & 7);
		    }
		    StreamError(0);
		    BSstate = BS_LOST;
		}
	    }
	    break;

	case BS_GAP:
	    count++;
	    if (bit == 2) {
		tprintf(2, "sample %d: skipped %d bits to mid-gap\n", time, count);
		BSstate = BS_LOST;
	    }
	    break;

	default:
	    assert(0);
	    break;
    }
}

// =========================================================================
// bit decoder

// sequentially the stream of transitions and turn them into a stream of
// bits.  of course, this routine must guard against illegal transitions
// that are bound to come up.

int
ClassifyTransition(uint32 time, uint32 duration)
{
    if (duration < 0.25f * pll_period)
	return -1;	// too short

    if (duration < 0.75f * pll_period)
	return 0;	// a half "zero" bit

    if (duration < 1.50f * pll_period)
	return 1;	// a "one" bit

    return 2;	// too long
}


// update the phase lock loop

// we want the phase detector to lock in on the right phase without
// being too responsive nor too slow.  this is just a guess.
// it is the amount the pll phase is adjusted at each bit cell time
// as a fraction of the error between expected and actual.
const float pll_bump = 0.15f;

void
PLL(int duration)
{
    const float phaseDiff = duration - pll_period;

    pll_period += phaseDiff * pll_bump;
    pll_period = MAX(min_samples_per_bit, pll_period);
    pll_period = MIN(max_samples_per_bit, pll_period);

    tprintf(4, "pll period = %f\n", pll_period);
}


void
DecodeBits(uint32 time)
{
    enum { DB_INIT=0, DB_HALF_BIT, DB_READY };
    static DBstate = DB_INIT;
    static int prevTime = 0;		// end of previous bit
    static int halfTime = 0;		// end of half bit
    int type;

    switch (DBstate) {

	case DB_INIT:
	    prevTime = time;
	    DBstate = DB_READY;
	    break;

	case DB_HALF_BIT:
	    type = ClassifyTransition(time, time-halfTime);
	    switch (type) {

		case -1:
		    if ((BSstate == BS_BYTE) && (BSbyteCount > 0))
			tprintf(1, "sample %d: Warning: runt pulse when short pulse expected\n", time);
		    Bit(time, -1);	// completed second short period
		    PLL(time-prevTime);	// correct for phase error
		    DBstate = DB_READY;
		    break;

		case 0:
		    Bit(time, 0);	// completed second short period
		    PLL(time-prevTime);	// correct for phase error
		    DBstate = DB_READY;
		    break;

		case 2:
		    if ((BSstate == BS_BYTE) && (BSbyteCount > 0))
			tprintf(1, "sample %d: Warning: really long pulse\n", time);
		    Bit(time, 2);
		    PLL(time-prevTime);	// correct for phase error
		    prevTime = time;
		    DBstate = DB_READY;
		    break;

		case 1:
		    if ((BSstate == BS_BYTE) && (BSbyteCount > 0))
			tprintf(1, "sample %d: long pulse when short pulse expected\n", time);
		    Bit(time, 1);
		    PLL(time-prevTime);	// correct for phase error
		    prevTime = time;
		    DBstate = DB_READY;
		    break;

		default:
		    assert(0);
		    break;
	    }
	    prevTime = time;
	    break;

	case DB_READY:
	    type = ClassifyTransition(time, time-prevTime);
	    switch (type) {

		case -1:
		    if ((BSstate == BS_BYTE) && (BSbyteCount > 0))
			tprintf(1, "sample %d: Warning: runt pulse\n", time);
		    Bit(time, -1);
		    halfTime = time;
		    DBstate = DB_READY;
		    break;

		case 0:
		    halfTime = time;
		    DBstate = DB_HALF_BIT;
		    break;

		case 2:
		    if (BSstate == BS_BYTE)
			tprintf(1, "sample %d: Warning: long pulse\n", time);
		    Bit(time, 2);
		    PLL(time-prevTime);	// correct for phase error
		    prevTime = time;
		    break;

		case 1:
		    Bit(time, 1);
		    PLL(time-prevTime);	// correct for phase error
		    prevTime = time;
		    break;

		default:
		    assert(0);
		    break;
	    }
	    break;

	default:
	    assert(0);
	    break;
    }
}


// =========================================================================
// find the duration of each flux zone

#if 1
// detect zero crossings directly
void
FindTransitions()
{
    bool     bFirst = true;	// first transition seen
    sample_t prevSamp;		// previous sample

    prevSamp = GETIN(0);
    for(uint32 nSamp=1; nSamp<expected_samples; nSamp++) {
	sample_t samp = GETIN(nSamp);
	if ((samp < 0) ^ (prevSamp < 0)) {
	    // a sign transition has occurred
	    if (!bFirst)
		DecodeBits(nSamp);
	    else
		bFirst = false;
	    prevSamp = samp;
	}
    }

    StreamDone();
}
#else
// find the alternating local minima and maxima and estimate the zero
// crossing to be midway between them.

this needs a lot more sophistication -- eg, look for a bit time or three
around the current point to get an estimate of the envelope of the wave
and use that when looking for peaks -- a deviation must be at least some
percentage of the min/max envelope spread.

void
FindTransitions()
{
    bool     bFirst = true;	// first transition seen
    bool     bUp    = false;
    sample_t prevSamp;		// previous sample
    sample_t prev_min   =  32000;
    uint32   prev_min_t = 0;		// time the min occurred
    sample_t prev_max   = -32000;
    uint32   prev_max_t = 0;		// time the max occurred
    float    prevIntercept = 0.0f;	// previous zero intercept

    prevSamp = GETIN(0);
    for(uint32 nSamp=1; nSamp<expected_samples; nSamp++) {
	sample_t samp = GETIN(nSamp);
	if (samp > prev_max) {
	    prev_max   = samp;
	    prev_max_t = nSamp;
	    bUp = true;
	} else if (samp < prev_min) {
	    prev_min   = samp;
	    prev_min_t = nSamp;
	    bUp = false;
	} else {
	    // look ahead a few samples to see if we just found the local min/max
	    bool atpeak = true;
	    for(int n=1; n<5; n++) {
		sample_t s = GETIN(nSamp + n);
		if ((s > prev_max) || (s < prev_min)) {
		    atpeak = false;
		    break;
		}
	    }
	    if (atpeak && (prev_max - prev_min > 2000)) {
		uint32 intercept = (prev_max_t + prev_min_t + 1) >> 1;
		if (bUp) {
printf("sample %d: detected max peak at %d\n", nSamp, prev_max);
		    prev_min = prev_max;
		} else {
printf("sample %d: detected min peak at %d\n", nSamp, prev_min);
		    prev_max = prev_min;
		}
		DecodeBits(intercept);
	    }
	}
    }

    StreamDone();
}
#endif


// =========================================================================
// main

void
usage(int code)
{
    FILE *f = (code == 0) ? stdout : stderr;

    fprintf(f, "Usage: dpwav2tap [-v [#]] [-x] [-o <outname>.tag] <inname>.wav\n");
    fprintf(f, "-v is report verbosity level\n");
    fprintf(f, "   with no -v, reporting is at a minimum;\n");
    fprintf(f, "   with no # specified, 1 is assumed;\n");
    fprintf(f, "   -v 2 through -v 4 provide increasing detail.\n");
    fprintf(f, "-o specifies a specific output filename;\n");
    fprintf(f, "   by default it is <inname>.tap\n");
    fprintf(f, "-x says to \"explode\" each record into a separate file,\n");
    fprintf(f, "   which are <outname>-###.tap\n");
    fprintf(f, "Version: August 24, 2005\n");
    exit(code);

#if 0
    // detailed -v information
    0 = just produce the output file and errors
    1 = errors, warnings, length of each decoded block
    2 = and each byte decoded
    3 = and each bit decoded
    4 = and PLL value
#endif
}


// parse the command line arguments
void
ParseArgs(int argc, char **argv)
{
    // set command line defaults
    opt_v     = 0;
    opt_x     = false;
    opt_x_num = 0;
    opt_ofn   = NULL;
    opt_ofmt  = OFMT_TAP;
    
#if 1
    if (argc < 2)	// we need at least one parameter
	usage(-1);
#endif

    for(int i=1; i<argc-1; i++) {

	if (strcmp(argv[i], "-v") == 0) {
	    // verbose reporting
	    opt_v = 1;  // at least
	    if (i+1 < argc-1) {
		// check for optional number
		opt_v = atol(argv[i+1]);
		if (opt_v == 0) {
		    // if next arg wasn't a number, atol returns 0
		    // however, if the user did "-v 0", we mess up in this assumption
		    opt_v = 1;
		} else
		    i++;	// skip numeric argument
	    }
	}

	else if (strcmp(argv[i], "-x") == 0) {
	    opt_x = true;
	}

	else if (strcmp(argv[i], "-o") == 0) {
	    if (i+1 < argc-1) {
		int len = strlen(argv[i+1]);
		opt_ofn = strdup(argv[i+1]);
		i++;
	    } else {
		fprintf(stderr, "Error: final argument is input filename, not part of -o specification\n");
		usage(-1);
	    }
	}

	else {
	    fprintf(stderr, "Error: unrecognized option '%s'\n", argv[i]);
	    usage(-1);
	}
    }

#if 1
    // one last parameter -- it must be the original wav file
    opt_ifn = strdup(argv[argc-1]);
#else
opt_v = 4;
opt_ifn = "c:\\jim\\datapoint\\dpwav2tap\\wav\\tstdis1.1_3-75.wav";
#endif

    // if -o wasn't specified, derive the output filename from the input filename
    if (opt_ofn == NULL) {
	// we just change the suffix of the input filename
	int len = strlen(opt_ifn);
	opt_ofn = (char *)malloc(len+4);  // +4 for appending optional .tap suffix
	assert(opt_ofn != NULL);
	strcpy(opt_ofn, opt_ifn);
	if ((len >= 4) && (stricmp(&opt_ifn[len-4], ".wav") == 0)) {
	    // input filename ends in ".wav"; replace it with .tap
	    strcpy(&opt_ofn[len-3], "tap");
	} else {
	    // input filename doesn't end like we'd expect
	    opt_ofn = strcat(opt_ifn, ".tap");
	}
    }

    // if -x is in effect, insert "-%03d" just before .tap
    // in order to generate serialized filenames.
    if (opt_x) {
	int len = strlen(opt_ofn);
	char *tmp = (char*)malloc(len+5);
	assert(tmp != NULL);
	strcpy(tmp, opt_ofn);
	if (stricmp(&opt_ofn[len-4], ".tap") == 0)
	    strcpy(&tmp[len-4], "-%03d.tap");
	else if (stricmp(&opt_ofn[len-4], ".bin") == 0)
	    strcpy(&tmp[len-4], "-%03d.bin");
#if SUPPORT_HEX
	else if (stricmp(&opt_ofn[len-4], ".hex") == 0)
	    strcpy(&tmp[len-4], "-%03d.hex");
#endif
	else
	    strcpy(&tmp[len], "-%03d");
	free(opt_ofn);
	opt_ofn = tmp;
    }

    {
	int len = strlen(opt_ofn);
	opt_ofmt = (stricmp(&opt_ofn[len-4], ".bin") == 0) ? OFMT_BIN
#if SUPPORT_HEX
		 : (stricmp(&opt_ofn[len-4], ".hex") == 0) ? OFMT_HEX
#endif
							   : OFMT_TAP;
    }

#if 0
    // report how command line was parsed
    printf("opt_v   = %d\n", opt_v);
    printf("opt_x   = %d\n", (int)opt_x);
    printf("opt_ofn = '%s'\n", opt_ofn);
    printf("opt_ifn = '%s'\n", opt_ifn);
    exit(0);
#endif
}


int
main(int argc, char **argv)
{
    // parse command line arguments
    ParseArgs(argc, argv);

    fIn = fopen(opt_ifn, "rb");
    if (fIn == NULL) {
	fprintf(stderr, "Error: couldn't open file '%s'\n", opt_ifn);
	exit(-1);
    }

    // make sure the wav file is OK
    CheckHeader();

    // initialize our work buffer
    // NB: the whole inbuffer complication is plumbing for a more
    //     sophisticated version of this program that looks forward
    //     and back in the sample stream.
    FillInbuffer();

    // process the file
    FindTransitions();

    fclose(fIn);

    return 0;
}