HRD Emulator in HTML5

Posted on March 21, 2012 by Moto

Just like MPEG-2 video uses VBV (Video Buffer Verifier), H.264 standard uses HRD (Hypothetical Reference Decoder) to define correctness of streams.

One of the verification provided by HRD is CPB buffer fullness. The idea is simple; feed H.264 stream to HRD and monitor the CPB fullness. Make sure the fullness never goes negative nor above the CPB buffer size. The importance of this buffer fullness requirement was explained here.

HRD defines the CPB fullness in somewhat counter intuitive way. To make it easy, I wrote an interactive HRD emulator in HTML5.

HRD Emulator (CBR)

The blue line the number of bits entered to HRD over time. The input rate is constant in CBR mode, i.e. the blue line is straight. The red line is the number of bits removed (= Decoded) from CPB. Thus, the CPB fullness is the blue line minus the red line for a given time.

Let’s select the VBR in Rate Control Mode.

HRD Emulator (VBR)

The blue line now has some periods where the input rate is zero. H.264 standard uses init_cpb_removal_delay to control the period; Bits for a picture N can enter HRD only after the picture decode time minus init_cpb_removal_delay.

To see this rule more clearly, check “Show init_cpb_removal_delay” option. When the last bit of picture N – 1 arrives before the picture N decoding time minus init_cpb_removal_delay, the input rate goes down to zero.

HRD (VBR) with init_cpb_removal_delay

The value of init_cpb_removal_delay also determines the initial decoding delay. It’s currently set to 50 ms. It means the first picture is decoded 50 ms after the first bit of the picture enters HRD.

The initial decoding delay determines the channel switching delay as discussed here. Can we reduce it to make channel switching faster?

HRD underflow due to the short initial delay

Not really. As shown above, reducing the value of init_cpb_removal_delay resulted in the underflow error (the red line crosses the blue line).

H.264 standard has introduced another parameter to allow the reduction of initial decoding delay at the expense of increased average bit rate. Here we have init_cpb_removal_delay_offset parameter. It’s added to init_cpb_removal_delay for every picture except the first one.

HRD with non-zero init_cpb_removal_delay_offset

H.264 HRD provides these parameters to encoders so that they can choose optimal bit stream configuration for each target devices.

Here are HTML and Java Script of the HRD emulator above.

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

61

62

63

64

65

66

67

68

69

70

<html>

<head>

<title>HRD Emulator</title>

#plot {

float : left

}

#control-container {

float : left

}

#widget {

float : right;

width : 10em

}

#heading {

float : left

}

</style>

</head>

</div>

</div>

<div class="heading">initial_cpb_removal_delay [<span id="showICRD"> </span> ms]</div>

<div>

</div>

<div class="heading">initial_cpb_removal_delay_offset [<span id="showICRDO"> </span> ms]</div>

<div>

</div>

<div>Rate Control Mode</div>

<div>

<input type="radio" name="mode" value="cbr" onClick="setMode('cbr'); update()" checked="checked" />CBR

</div>

<div>

<input type="radio" name="mode" value="vbr" onClick="setMode('vbr'); update()" />VBR

</div>

<div>

<input type="checkbox" id="isICRDVisible" onClick="update()"/>Show init_cpb_removal_delay

</div>

<div>

<input type="checkbox" id="isICRDOVisible" onClick="update()"/>Show init_cpb_removal_delay_offset

</div>

</div>

</body>

</html>

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

61

62

63

64

65

66

67

68

69

70

71

72

73

74

75

76

77

78

79

80

81

82

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

// Save this as "script.js"

var canvas;

var picture = new Array();

var pictureDuration = 900900;

var initial_cpb_removal_delay = 900900;

var initial_cpb_removal_delay_offset = 90090;

var cbr_flag = true;

function init()

{

canvas = document.getElementById('canvas');

resetStream();

update();

}

function resetStream()

{

var bit_rate = document.getElementById('bitRate').value * 1000;

var averagePicSize = pictureDuration / 27000000 * bit_rate / 8;

for(var i = 0; i < 10; i++){

picture[i] = averagePicSize + (Math.random() - 0.5) * averagePicSize;

}

function tickToX(tick)

{

return tick / 900900 * 64;

}

function levelToY(level)

{

return canvas.height - level / 50;

}

function drawDecodePictures(ctx)

{

ctx.strokeStyle = 'rgba(255, 0, 0, 255)';

ctx.beginPath();

ctx.moveTo(0, levelToY(0));

ctx.lineTo(tickToX(initial_cpb_removal_delay), levelToY(0));

var prevX = tickToX(initial_cpb_removal_delay)

var prevLevel = 0;

for(var i = 0; i < 50; i++){

var level = prevLevel + picture[i];

var x1 = prevX;

var x2 = prevX + tickToX(pictureDuration)

var y = levelToY(level);

ctx.lineTo(x1, y);

ctx.lineTo(x2, y);

prevLevel = level;

prevX = x2

}

ctx.stroke();

}

function levelFromPeriod(period, bitRate)

{

return period / 27000000.0 * bitRate / 8;

}

function drawInputData(ctx)

{

var t_af = new Array();

var t_ai = new Array();

var t_ai_earliest = new Array();

var t_rn = new Array();

var bit_rate = document.getElementById('bitRate').value * 1000;

t_ai[0] = 0;

t_af[0] = picture[0] * 8 * 27000000 / bit_rate;

t_rn[0] = initial_cpb_removal_delay;

for(var n = 1; n < picture.length; n++)

{

t_rn[n] = t_rn[n - 1] + pictureDuration;

if(cbr_flag == true)

{

t_ai[n] = t_af[n - 1];

}

else

{

t_ai_earliest = t_rn[n] - (initial_cpb_removal_delay + initial_cpb_removal_delay_offset);

t_ai[n] = Math.max(t_af[n - 1], t_ai_earliest);

}

t_af[n] = t_ai[n] + picture[n] * 8 * 27000000 / bit_rate;

}

ctx.strokeStyle = 'rgba(0, 0, 255, 255)';

ctx.beginPath();

ctx.moveTo(tickToX(0), levelToY(0));

var level = 0;

for(var n = 0; n < picture.length; n++)

{

ctx.lineTo(tickToX(t_ai[n]), levelToY(level));

level += picture[n];

ctx.lineTo(tickToX(t_af[n]), levelToY(level));

}

ctx.stroke();

level = 0;

for(var n = 0; n < picture.length; n++)

{

var isICRDVisible = document.getElementById("isICRDVisible").checked;

if(isICRDVisible){

// draw initial_cpb_removal_delay period

var icrd_x1 = tickToX(t_rn[n] - initial_cpb_removal_delay)

var icrd_x2 = tickToX(t_rn[n])

var icrd_y = levelToY(level) - 4

ctx.strokeStyle = 'rgba(0, 0, 0, 255)';

ctx.beginPath();

ctx.moveTo(icrd_x1, icrd_y);

ctx.lineTo(icrd_x2, icrd_y);

ctx.stroke();

ctx.beginPath();

ctx.moveTo(icrd_x1 + 2, icrd_y - 2);

ctx.lineTo(icrd_x1, icrd_y);

ctx.lineTo(icrd_x1 + 2, icrd_y + 2);

ctx.stroke();

ctx.beginPath();

ctx.moveTo(icrd_x2 - 2, icrd_y - 2);

ctx.lineTo(icrd_x2, icrd_y);

ctx.lineTo(icrd_x2 - 2, icrd_y + 2);

ctx.stroke();

}

var isICRDOVisible = document.getElementById("isICRDOVisible").checked;

if(isICRDOVisible){

// draw initial_cpb_removal_delay period

var icrdo_x1 = tickToX(t_rn[n] - initial_cpb_removal_delay - initial_cpb_removal_delay_offset)

var icrdo_x2 = icrd_x1

var icrdo_y = levelToY(level) - 4

ctx.strokeStyle = 'rgba(0, 0, 0, 255)';

ctx.beginPath();

ctx.moveTo(icrdo_x1, icrdo_y);

ctx.lineTo(icrdo_x2, icrdo_y);

ctx.stroke();

ctx.beginPath();

ctx.moveTo(icrdo_x1 + 2, icrdo_y - 2);

ctx.lineTo(icrdo_x1, icrdo_y);

ctx.lineTo(icrdo_x1 + 2, icrdo_y + 2);

ctx.stroke();

ctx.beginPath();

ctx.moveTo(icrdo_x2 - 2, icrdo_y - 2);

ctx.lineTo(icrdo_x2, icrdo_y);

ctx.lineTo(icrdo_x2 - 2, icrdo_y + 2);

ctx.stroke();

}

level += picture[n];

}

function setMode(mode)

{

cbr_flag = mode == "cbr";

}

function updateUI(uiName, valueName)

{

document.getElementById(uiName).innerHTML = document.getElementById(valueName).value

}

function update()

{

var ctx = canvas.getContext('2d');

ctx.fillStyle = 'rgba(255, 255, 255, 255)';

ctx.fillRect(0, 0, canvas.width, canvas.height);

updateUI("showBitRate", "bitRate");

updateUI("showICRD", "initial_cpb_removal_delay");

updateUI("showICRDO", "initial_cpb_removal_delay_offset");

initial_cpb_removal_delay = document.getElementById('initial_cpb_removal_delay').value * 27000;

initial_cpb_removal_delay_offset = document.getElementById('initial_cpb_removal_delay_offset').value * 27000;

drawDecodePictures(ctx);

drawInputData(ctx);

};

相关阅读:
448-查找数组中消失的所有数字
977 -排序数组的正方形
爬虫小总结
增量式爬虫
分布式爬虫
CrawlSpider：类，Spider的一个子类
中间件
中间件
scrapy图片数据爬取之ImagesPipeline
scrapy五大核心组件

原文地址：https://www.cnblogs.com/xkfz007/p/6391491.html