add support for spectral extension

Originally committed as revision 15812 to svn://svn.ffmpeg.org/ffmpeg/trunk

5 files changed, 272 insertions, 17 deletions

diff --git a/libavcodec/ac3dec.c b/libavcodec/ac3dec.c
index 3e8b0b5afc..fb6fc61340 100644
--- a/libavcodec/ac3dec.c
+++ b/libavcodec/ac3dec.c
@@ -818,14 +818,86 @@ static int decode_audio_block(AC3DecodeContext *s, int blk)
 
     /* spectral extension strategy */
     if (s->eac3 && (!blk || get_bits1(gbc))) {
-        if (get_bits1(gbc)) {
-            av_log_missing_feature(s->avctx, "Spectral extension", 1);
-            return -1;
+        s->spx_in_use[blk] = get_bits1(gbc);
+        if (s->spx_in_use[blk]) {
+            int begf, endf;
+            int spx_end_subband;
+
+            /* determine which channels use spx */
+            if (s->channel_mode == AC3_CHMODE_MONO) {
+                s->channel_in_spx[1] = 1;
+                s->spx_coords_exist[1] = 0;
+            } else {
+                for (ch = 1; ch <= fbw_channels; ch++) {
+                    s->channel_in_spx[ch] = get_bits1(gbc);
+                    s->spx_coords_exist[ch] = 0;
+                }
+            }
+
+            s->spx_copy_start_freq = get_bits(gbc, 2) * 12 + 25;
+            begf = get_bits(gbc, 3);
+            endf = get_bits(gbc, 3);
+            s->spx_start_subband = begf < 6 ? begf+2 : 2*begf-3;
+            spx_end_subband      = endf < 4 ? endf+5 : 2*endf+3;
+            s->num_spx_subbands  = spx_end_subband - s->spx_start_subband;
+            s->spx_start_freq    = s->spx_start_subband * 12 + 25;
+            s->spx_end_freq      = spx_end_subband      * 12 + 25;
+
+            decode_band_structure(gbc, blk, s->eac3, 0,
+                                  s->spx_start_subband, spx_end_subband,
+                                  ff_eac3_default_spx_band_struct,
+                                  s->spx_band_struct, NULL, &s->num_spx_bands,
+                                  s->spx_band_sizes);
+        } else {
+            for (ch = 1; ch <= fbw_channels; ch++) {
+                s->channel_in_spx[ch] = 0;
+                s->first_spx_coords[ch] = 1;
+            }
         }
-        /* TODO: parse spectral extension strategy info */
+    } else {
+        s->spx_in_use[blk] = blk ? s->spx_in_use[blk-1] : 0;
     }
 
-    /* TODO: spectral extension coordinates */
+    /* spectral extension coordinates */
+    if (s->spx_in_use[blk]) {
+        for (ch = 1; ch <= fbw_channels; ch++) {
+            if (s->channel_in_spx[ch]) {
+                if (s->first_spx_coords[ch] || get_bits1(gbc)) {
+                    int bin, spx_blend;
+                    int master_spx_coord;
+                    s->first_spx_coords[ch] = 0;
+                    s->spx_coords_exist[ch] = 1;
+                    spx_blend = get_bits(gbc, 5) << 18;
+                    master_spx_coord = get_bits(gbc, 2) * 3;
+                    bin = s->spx_start_freq;
+                    for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
+                        int spx_coord_exp, spx_coord_mant;
+
+                        /* calculate blending factors */
+                        int bandsize = s->spx_band_sizes[bnd];
+                        int nratio = (((bin + (bandsize >> 1)) << 23) / s->spx_end_freq) - spx_blend;
+                        nratio = av_clip(nratio, 0, INT24_MAX);
+                        s->spx_noise_blend [ch][bnd] = ff_sqrt((            nratio) << 8) * M_SQRT_POW2_15;
+                        s->spx_signal_blend[ch][bnd] = ff_sqrt((INT24_MAX - nratio) << 8) * M_SQRT_POW2_15;
+                        bin += bandsize;
+
+                        /* decode spx coordinates */
+                        spx_coord_exp  = get_bits(gbc, 4);
+                        spx_coord_mant = get_bits(gbc, 2);
+                        if (spx_coord_exp == 15)
+                            s->spx_coords[ch][bnd] = spx_coord_mant << 26;
+                        else
+                            s->spx_coords[ch][bnd] = (spx_coord_mant + 4) << 25;
+                        s->spx_coords[ch][bnd] >>= (spx_coord_exp + master_spx_coord);
+                    }
+                } else {
+                    s->spx_coords_exist[ch] = 0;
+                }
+            } else {
+                s->first_spx_coords[ch] = 1;
+            }
+        }
+    }
 
     /* coupling strategy */
     if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
@@ -862,9 +934,12 @@ static int decode_audio_block(AC3DecodeContext *s, int blk)
                 s->phase_flags_in_use = get_bits1(gbc);
 
             /* coupling frequency range */
-            /* TODO: modify coupling end freq if spectral extension is used */
             cpl_start_subband = get_bits(gbc, 4);
+            if (s->spx_in_use[blk]) {
+                cpl_end_subband = s->spx_start_subband - 1;
+            } else {
             cpl_end_subband   = get_bits(gbc, 4) + 3;
+            }
             s->num_cpl_subbands = cpl_end_subband - cpl_start_subband;
             if (s->num_cpl_subbands < 0) {
                 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d > %d)\n",
@@ -939,8 +1014,14 @@ static int decode_audio_block(AC3DecodeContext *s, int blk)
     if (channel_mode == AC3_CHMODE_STEREO) {
         if ((s->eac3 && !blk) || get_bits1(gbc)) {
             s->num_rematrixing_bands = 4;
-            if(cpl_in_use && s->start_freq[CPL_CH] <= 61)
+            if (cpl_in_use) {
+                if (s->start_freq[CPL_CH] <= 61)
                 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
+            } else if (s->spx_in_use[blk]) {
+                if (s->spx_start_freq <= 61)
+                    s->num_rematrixing_bands -= 1 + (s->spx_start_freq <= 37) +
+                                                    (s->spx_start_freq <= 25);
+            }
             for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
                 s->rematrixing_flags[bnd] = get_bits1(gbc);
         } else if (!blk) {
@@ -965,6 +1046,8 @@ static int decode_audio_block(AC3DecodeContext *s, int blk)
             int prev = s->end_freq[ch];
             if (s->channel_in_cpl[ch])
                 s->end_freq[ch] = s->start_freq[CPL_CH];
+            else if (s->channel_in_spx[ch])
+                s->end_freq[ch] = s->spx_start_freq;
             else {
                 int bandwidth_code = get_bits(gbc, 6);
                 if (bandwidth_code > 60) {
@@ -1155,12 +1238,12 @@ static int decode_audio_block(AC3DecodeContext *s, int blk)
 
     /* TODO: generate enhanced coupling coordinates and uncouple */
 
-    /* TODO: apply spectral extension */
-
     /* recover coefficients if rematrixing is in use */
     if(s->channel_mode == AC3_CHMODE_STEREO)
         do_rematrixing(s);
 
+    ff_eac3_apply_spectral_extension(s);
+
     /* apply scaling to coefficients (headroom, dynrng) */
     for(ch=1; ch<=s->channels; ch++) {
         float gain = s->mul_bias / 4194304.0f;
diff --git a/libavcodec/ac3dec.h b/libavcodec/ac3dec.h
index 617080910b..7d34837e0c 100644
--- a/libavcodec/ac3dec.h
+++ b/libavcodec/ac3dec.h
@@ -42,6 +42,11 @@
 #define AC3_MAX_COEFS   256
 #define AC3_BLOCK_SIZE  256
 #define MAX_BLOCKS        6
+#define SPX_MAX_BANDS    17
+
+#define INT24_MIN -8388608
+#define INT24_MAX  8388607
+#define M_SQRT_POW2_15 181
 
 typedef struct {
     AVCodecContext *avctx;                  ///< parent context
@@ -88,6 +93,26 @@ typedef struct {
     int cpl_coords[AC3_MAX_CHANNELS][18];   ///< coupling coordinates                   (cplco)
 ///@}
 
+///@defgroup spx spectral extension
+///@{
+    int spx_in_use[MAX_BLOCKS];             ///< spectral extension in use              (spxinu)
+    int channel_in_spx[AC3_MAX_CHANNELS];   ///< channel in spectral extension          (chinspx)
+    int spx_atten_code[AC3_MAX_CHANNELS];   ///< spx attenuation code                   (spxattencod)
+    int spx_coords_exist[AC3_MAX_CHANNELS]; ///< indicates if a channel has spx coords  (spxcoe)
+    int spx_start_subband;                  ///< spx beginning frequency band           (spxbegf)
+    int spx_start_freq;                     ///< spx start frequency bin
+    int spx_end_freq;                       ///< spx end frequency bin
+    int spx_copy_start_freq;                ///< spx starting frequency for copying     (copystartmant)
+    int num_spx_subbands;                   ///< number of spectral extension subbands
+    int num_spx_bands;                      ///< number of spectral extension bands     (nspxbnds)
+    uint8_t spx_band_struct[SPX_MAX_BANDS]; ///< spectral extension band structure      (spxbndstrc)
+    int spx_band_sizes[SPX_MAX_BANDS];      ///< number of bins in each band            (spxbndsztab)
+    int first_spx_coords[AC3_MAX_CHANNELS]; ///< first spx coordinates states           (firstspxcos)
+    int spx_noise_blend[AC3_MAX_CHANNELS][SPX_MAX_BANDS];   ///< spx noise blending factor  (nblendfact)
+    int spx_signal_blend[AC3_MAX_CHANNELS][SPX_MAX_BANDS];  ///< spx signal blending factor (sblendfact)
+    int spx_coords[AC3_MAX_CHANNELS][SPX_MAX_BANDS];    ///< spectral extension coordinates (spxco)
+///@}
+
 ///@defgroup aht adaptive hybrid transform
     int channel_uses_aht[AC3_MAX_CHANNELS];                         ///< channel AHT in use (chahtinu)
     int pre_mantissa[AC3_MAX_CHANNELS][AC3_MAX_COEFS][MAX_BLOCKS];  ///< pre-IDCT mantissas
@@ -179,4 +204,6 @@ int ff_eac3_parse_header(AC3DecodeContext *s);
  */
 void ff_eac3_decode_transform_coeffs_aht_ch(AC3DecodeContext *s, int ch);
 
+void ff_eac3_apply_spectral_extension(AC3DecodeContext *s);
+
 #endif /* AVCODEC_AC3DEC_H */
diff --git a/libavcodec/ac3dec_data.c b/libavcodec/ac3dec_data.c
index a70c8fca45..fb21888c78 100644
--- a/libavcodec/ac3dec_data.c
+++ b/libavcodec/ac3dec_data.c
@@ -1127,6 +1127,52 @@ const uint8_t ff_eac3_frm_expstr[32][6] = {
 const uint8_t ff_eac3_default_cpl_band_struct[18] =
 { 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 1, 0, 1, 1, 1, 1, 1 };
 
+ /**
+ * Table E2.15 Default Spectral Extension Banding Structure
+ */
+const uint8_t ff_eac3_default_spx_band_struct[17] =
+{ 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1 };
+
+/**
+ * Table E.25: Spectral Extension Attenuation Table
+ * 24-bit fixed-point version of the floating-point table in the specification.
+ * ff_eac3_spx_atten_tab[code][bin]=lrint(pow(1<<(bin+1),(code+1)/-15.0)*(1<<23));
+ */
+const int32_t ff_eac3_spx_atten_tab[32][3] = {
+    { 8009792, 7648083, 7302707 },
+    { 7648083, 6972929, 6357376 },
+    { 7302707, 6357376, 5534417 },
+    { 6972929, 5796163, 4817990 },
+    { 6658043, 5284492, 4194304 },
+    { 6357376, 4817990, 3651354 },
+    { 6070287, 4392670, 3178688 },
+    { 5796163, 4004896, 2767209 },
+    { 5534417, 3651354, 2408995 },
+    { 5284492, 3329021, 2097152 },
+    { 5045853, 3035144, 1825677 },
+    { 4817990, 2767209, 1589344 },
+    { 4600417, 2522926, 1383604 },
+    { 4392670, 2300209, 1204498 },
+    { 4194304, 2097152, 1048576 },
+    { 4004896, 1912021,  912838 },
+    { 3824041, 1743232,  794672 },
+    { 3651354, 1589344,  691802 },
+    { 3486464, 1449041,  602249 },
+    { 3329021, 1321123,  524288 },
+    { 3178688, 1204498,  456419 },
+    { 3035144, 1098167,  397336 },
+    { 2898081, 1001224,  345901 },
+    { 2767209,  912838,  301124 },
+    { 2642246,  832255,  262144 },
+    { 2522926,  758786,  228210 },
+    { 2408995,  691802,  198668 },
+    { 2300209,  630732,  172951 },
+    { 2196335,  575052,  150562 },
+    { 2097152,  524288,  131072 },
+    { 2002448,  478005,  114105 },
+    { 1912021,  435808,   99334 }
+};
+
 /**
  * Table of bin locations for rematrixing bands
  * reference: Section 7.5.2 Rematrixing : Frequency Band Definitions
diff --git a/libavcodec/ac3dec_data.h b/libavcodec/ac3dec_data.h
index 486d04a40b..73aae1e178 100644
--- a/libavcodec/ac3dec_data.h
+++ b/libavcodec/ac3dec_data.h
@@ -34,6 +34,8 @@ extern const int8_t  ff_eac3_gaq_remap_2_4_b[9][2];
 extern const int16_t (* const ff_eac3_mantissa_vq[8])[6];
 extern const uint8_t ff_eac3_frm_expstr[32][6];
 extern const uint8_t ff_eac3_default_cpl_band_struct[18];
+extern const uint8_t ff_eac3_default_spx_band_struct[17];
+extern const int32_t ff_eac3_spx_atten_tab[32][3];
 
 extern const uint8_t ff_ac3_rematrix_band_tab[5];
 
diff --git a/libavcodec/eac3dec.c b/libavcodec/eac3dec.c
index 0b10b41848..58376f2929 100644
--- a/libavcodec/eac3dec.c
+++ b/libavcodec/eac3dec.c
@@ -36,6 +36,104 @@ typedef enum {
 
 #define EAC3_SR_CODE_REDUCED  3
 
+
+void ff_eac3_apply_spectral_extension(AC3DecodeContext *s)
+{
+    int bin, bnd, ch, i;
+    int wrapflag[SPX_MAX_BANDS]={0,}, num_copy_sections, copy_sizes[SPX_MAX_BANDS];
+    int rms_energy[SPX_MAX_BANDS];
+
+    /* Set copy index mapping table. Set wrap flags to apply a notch filter at
+       wrap points later on. */
+    bin = s->spx_copy_start_freq;
+    num_copy_sections = 0;
+    for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
+        int bandsize = s->spx_band_sizes[bnd];
+        if ((bin + bandsize) > s->spx_start_freq) {
+            copy_sizes[num_copy_sections++] = bin - s->spx_copy_start_freq;
+            bin = s->spx_copy_start_freq;
+            wrapflag[bnd] = 1;
+        }
+        for (i = 0; i < bandsize; i++) {
+            if (bin == s->spx_start_freq) {
+                copy_sizes[num_copy_sections++] = bin - s->spx_copy_start_freq;
+                bin = s->spx_copy_start_freq;
+            }
+            bin++;
+        }
+    }
+    copy_sizes[num_copy_sections++] = bin - s->spx_copy_start_freq;
+
+    for (ch = 1; ch <= s->fbw_channels; ch++) {
+        if (!s->channel_in_spx[ch])
+            continue;
+
+        /* Copy coeffs from normal bands to extension bands */
+        bin = s->spx_start_freq;
+        for (bnd = 0; bnd < num_copy_sections; bnd++) {
+            memcpy(&s->fixed_coeffs[ch][bin],
+                   &s->fixed_coeffs[ch][s->spx_copy_start_freq],
+                   copy_sizes[bnd]*sizeof(int));
+            bin += copy_sizes[bnd];
+        }
+
+        /* Calculate RMS energy for each SPX band. */
+        bin = s->spx_start_freq;
+        for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
+            int bandsize = s->spx_band_sizes[bnd];
+            int64_t accum = 0;
+            for (i = 0; i < bandsize; i++) {
+                int64_t coeff = s->fixed_coeffs[ch][bin++];
+                accum += coeff * coeff;
+            }
+            rms_energy[bnd] = ff_sqrt((accum >> 15) / bandsize) * M_SQRT_POW2_15;
+        }
+
+        /* Apply a notch filter at transitions between normal and extension
+           bands and at all wrap points. */
+        if (s->spx_atten_code[ch] >= 0) {
+            const int32_t *atten_tab = ff_eac3_spx_atten_tab[s->spx_atten_code[ch]];
+            /* apply notch filter at baseband / extension region border */
+            bin = s->spx_start_freq - 2;
+            for (i = 0; i < 5; i++) {
+                s->fixed_coeffs[ch][bin] = ((int64_t)atten_tab[2-abs(i-2)] *
+                        (int64_t)s->fixed_coeffs[ch][bin]) >> 23;
+                bin++;
+            }
+            /* apply notch at all other wrap points */
+            bin += s->spx_band_sizes[0];
+            for (bnd = 1; bnd < s->num_spx_bands; bnd++) {
+                if (wrapflag[bnd]) {
+                    bin -= 5;
+                    for (i = 0; i < 5; i++) {
+                        s->fixed_coeffs[ch][bin] = (atten_tab[2-abs(i-2)] *
+                                (int64_t)s->fixed_coeffs[ch][bin]) >> 23;
+                        bin++;
+                    }
+                }
+                bin += s->spx_band_sizes[bnd];
+            }
+        }
+
+        /* Apply noise-blended coefficient scaling based on previously
+           calculated RMS energy, blending factors, and SPX coordinates for
+           each band. */
+        bin = s->spx_start_freq;
+        for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
+            int64_t nscale, sscale, spxco;
+            nscale = (s->spx_noise_blend [ch][bnd] * rms_energy[bnd]) >> 23;
+            nscale = (nscale * 14529495) >> 23;
+            sscale = s->spx_signal_blend[ch][bnd];
+            spxco  = s->spx_coords[ch][bnd];
+            for (i = 0; i < s->spx_band_sizes[bnd]; i++) {
+                int64_t noise  = (nscale * (((int)av_lfg_get(&s->dith_state))>>8)) >> 23;
+                int64_t signal = (sscale * s->fixed_coeffs[ch][bin]) >> 23;
+                s->fixed_coeffs[ch][bin++] = ((noise + signal) * spxco) >> 23;
+            }
+        }
+    }
+}
+
 /** lrint(M_SQRT2*cos(2*M_PI/12)*(1<<23)) */
 #define COEFF_0 10273905LL
 
@@ -459,14 +557,12 @@ int ff_eac3_parse_header(AC3DecodeContext *s)
     }
 
     /* spectral extension attenuation data */
-    if (parse_spx_atten_data) {
-        av_log_missing_feature(s->avctx, "Spectral extension attenuation", 1);
-        for (ch = 1; ch <= s->fbw_channels; ch++) {
-            if (get_bits1(gbc)) { // channel has spx attenuation
-                skip_bits(gbc, 5); // skip spx attenuation code
-            }
-        }
-    }
+    for (ch = 1; ch <= s->fbw_channels; ch++) {
+        if (parse_spx_atten_data && get_bits1(gbc))
+            s->spx_atten_code[ch] = get_bits(gbc, 5);
+        else
+            s->spx_atten_code[ch] = -1;
+     }
 
     /* block start information */
     if (s->num_blocks > 1 && get_bits1(gbc)) {
@@ -480,6 +576,7 @@ int ff_eac3_parse_header(AC3DecodeContext *s)
 
     /* syntax state initialization */
     for (ch = 1; ch <= s->fbw_channels; ch++) {
+        s->first_spx_coords[ch] = 1;
         s->first_cpl_coords[ch] = 1;
     }
     s->first_cpl_leak = 1;