File size: 21,258 Bytes
06555b5 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 |
import math
from enum import Enum
from typing import Dict, List, Optional, Tuple
from aiortc.utils import uint32_add, uint32_gt
BURST_DELTA_THRESHOLD_MS = 5
# overuse detector
MAX_ADAPT_OFFSET_MS = 15
MIN_NUM_DELTAS = 60
# overuse estimator
DELTA_COUNTER_MAX = 1000
MIN_FRAME_PERIOD_HISTORY_LENGTH = 60
# abs-send-time estimator
INTER_ARRIVAL_SHIFT = 26
TIMESTAMP_GROUP_LENGTH_MS = 5
TIMESTAMP_TO_MS = 1000.0 / (1 << INTER_ARRIVAL_SHIFT)
class BandwidthUsage(Enum):
NORMAL = 0
UNDERUSING = 1
OVERUSING = 2
class RateControlState(Enum):
HOLD = 0
INCREASE = 1
DECREASE = 2
class AimdRateControl:
def __init__(self) -> None:
self.avg_max_bitrate_kbps = None
self.var_max_bitrate_kbps = 0.4
self.current_bitrate = 30000000
self.current_bitrate_initialized = False
self.first_estimated_throughput_time: Optional[int] = None
self.last_change_ms: Optional[int] = None
self.near_max = False
self.latest_estimated_throughput = 30000000
self.rtt = 200
self.state = RateControlState.HOLD
def feedback_interval(self) -> int:
return 500
def set_estimate(self, bitrate: int, now_ms: int) -> None:
"""
For testing purposes.
"""
self.current_bitrate = self._clamp_bitrate(bitrate, bitrate)
self.current_bitrate_initialized = True
self.last_change_ms = now_ms
def update(
self,
bandwidth_usage: BandwidthUsage,
estimated_throughput: Optional[int],
now_ms: int,
) -> Optional[int]:
if not self.current_bitrate_initialized and estimated_throughput is not None:
if self.first_estimated_throughput_time is None:
self.first_estimated_throughput_time = now_ms
elif now_ms - self.first_estimated_throughput_time > 3000:
self.current_bitrate = estimated_throughput
self.current_bitrate_initialized = True
# wait for initialisation or overuse
if (
not self.current_bitrate_initialized
and bandwidth_usage != BandwidthUsage.OVERUSING
):
return None
# update state
if (
bandwidth_usage == BandwidthUsage.NORMAL
and self.state == RateControlState.HOLD
):
self.last_change_ms = now_ms
self.state = RateControlState.INCREASE
elif bandwidth_usage == BandwidthUsage.OVERUSING:
self.state = RateControlState.DECREASE
elif bandwidth_usage == BandwidthUsage.UNDERUSING:
self.state = RateControlState.HOLD
# helper variables
new_bitrate = self.current_bitrate
if estimated_throughput is not None:
self.latest_estimated_throughput = estimated_throughput
else:
estimated_throughput = self.latest_estimated_throughput
estimated_throughput_kbps = estimated_throughput / 1000
# update bitrate
if self.state == RateControlState.INCREASE:
# if the estimated throughput increases significantly,
# clear estimated max throughput
if self.avg_max_bitrate_kbps is not None:
sigma_kbps = math.sqrt(
self.var_max_bitrate_kbps * self.avg_max_bitrate_kbps
)
if (
estimated_throughput_kbps
>= self.avg_max_bitrate_kbps + 3 * sigma_kbps
):
self.near_max = False
self.avg_max_bitrate_kbps = None
# we use additive or multiplicative rate increase depending on whether
# we are close to the maximum throughput
if self.near_max:
new_bitrate += self._additive_rate_increase(self.last_change_ms, now_ms)
else:
new_bitrate += self._multiplicative_rate_increase(
new_bitrate, self.last_change_ms, now_ms
)
self.last_change_ms = now_ms
elif self.state == RateControlState.DECREASE:
# if the estimated throughput drops significantly,
# clear estimated max throughput
if self.avg_max_bitrate_kbps is not None:
sigma_kbps = math.sqrt(
self.var_max_bitrate_kbps * self.avg_max_bitrate_kbps
)
if (
estimated_throughput_kbps
< self.avg_max_bitrate_kbps - 3 * sigma_kbps
):
self.avg_max_bitrate_kbps = None
self._update_max_throughput_estimate(estimated_throughput_kbps)
self.near_max = True
new_bitrate = round(0.85 * estimated_throughput)
self.last_change_ms = now_ms
self.state = RateControlState.HOLD
self.current_bitrate = self._clamp_bitrate(new_bitrate, estimated_throughput)
return self.current_bitrate
def _additive_rate_increase(self, last_ms: int, now_ms: int) -> int:
return int((now_ms - last_ms) * self._near_max_rate_increase() / 1000)
def _clamp_bitrate(self, new_bitrate: int, estimated_throughput: int) -> int:
max_bitrate = max(int(1.5 * estimated_throughput) + 10000, self.current_bitrate)
return min(new_bitrate, max_bitrate)
def _multiplicative_rate_increase(
self, new_bitrate: int, last_ms: int, now_ms: int
) -> int:
alpha = 1.08
if last_ms is not None:
elapsed_ms = min(now_ms - last_ms, 1000)
alpha = pow(alpha, elapsed_ms / 1000)
return int(max((alpha - 1) * new_bitrate, 1000))
def _near_max_rate_increase(self) -> int:
bits_per_frame = self.current_bitrate / 30
packets_per_frame = math.ceil(bits_per_frame / (8 * 1200))
avg_packet_size_bits = bits_per_frame / packets_per_frame
response_time = self.rtt + 100
return max(4000, int((avg_packet_size_bits * 1000) / response_time))
def _update_max_throughput_estimate(self, estimated_throughput_kbps) -> None:
alpha = 0.05
if self.avg_max_bitrate_kbps is None:
self.avg_max_bitrate_kbps = estimated_throughput_kbps
else:
self.avg_max_bitrate_kbps = (
1 - alpha
) * self.avg_max_bitrate_kbps + alpha * estimated_throughput_kbps
norm = max(1, self.avg_max_bitrate_kbps)
self.var_max_bitrate_kbps = (1 - alpha) * self.var_max_bitrate_kbps + alpha * (
(self.avg_max_bitrate_kbps - estimated_throughput_kbps) ** 2
) / norm
self.var_max_bitrate_kbps = max(0.4, min(self.var_max_bitrate_kbps, 2.5))
class TimestampGroup:
def __init__(self, timestamp: Optional[int] = None) -> None:
self.arrival_time: Optional[int] = None
self.first_timestamp = timestamp
self.last_timestamp = timestamp
self.size = 0
class InterArrivalDelta:
def __init__(self, timestamp: int, arrival_time: int, size: int) -> None:
self.timestamp = timestamp
self.arrival_time = arrival_time
self.size = size
class InterArrival:
"""
Inter-arrival time and size filter.
Adapted from the webrtc.org codebase.
"""
def __init__(self, group_length: int, timestamp_to_ms: float) -> None:
self.group_length = group_length
self.timestamp_to_ms = timestamp_to_ms
self.current_group: Optional[TimestampGroup] = None
self.previous_group: Optional[TimestampGroup] = None
def compute_deltas(
self, timestamp: int, arrival_time: int, packet_size: int
) -> Optional[InterArrivalDelta]:
deltas = None
if self.current_group is None:
self.current_group = TimestampGroup(timestamp)
elif self.packet_out_of_order(timestamp):
return deltas
elif self.new_timestamp_group(timestamp, arrival_time):
if self.previous_group is not None:
deltas = InterArrivalDelta(
timestamp=uint32_add(
self.current_group.last_timestamp,
-self.previous_group.last_timestamp,
),
arrival_time=(
self.current_group.arrival_time
- self.previous_group.arrival_time
),
size=self.current_group.size - self.previous_group.size,
)
# shift groups
self.previous_group = self.current_group
self.current_group = TimestampGroup(timestamp=timestamp)
elif uint32_gt(timestamp, self.current_group.last_timestamp):
self.current_group.last_timestamp = timestamp
self.current_group.size += packet_size
self.current_group.arrival_time = arrival_time
return deltas
def belongs_to_burst(self, timestamp: int, arrival_time: int) -> bool:
timestamp_delta = uint32_add(timestamp, -self.current_group.last_timestamp)
timestamp_delta_ms = round(self.timestamp_to_ms * timestamp_delta)
arrival_time_delta = arrival_time - self.current_group.arrival_time
return timestamp_delta_ms == 0 or (
(arrival_time_delta - timestamp_delta_ms) < 0
and arrival_time_delta <= BURST_DELTA_THRESHOLD_MS
)
def new_timestamp_group(self, timestamp: int, arrival_time: int) -> bool:
if self.belongs_to_burst(timestamp, arrival_time):
return False
else:
timestamp_delta = uint32_add(timestamp, -self.current_group.first_timestamp)
return timestamp_delta > self.group_length
def packet_out_of_order(self, timestamp: int) -> bool:
timestamp_delta = uint32_add(timestamp, -self.current_group.first_timestamp)
return timestamp_delta >= 0x80000000
class OveruseDetector:
"""
Bandwidth overuse detector.
Adapted from the webrtc.org codebase.
"""
def __init__(self) -> None:
self.hypothesis = BandwidthUsage.NORMAL
self.last_update_ms: Optional[int] = None
self.k_up = 0.0087
self.k_down = 0.039
self.overuse_counter = 0
self.overuse_time: Optional[float] = None
self.overuse_time_threshold = 10
self.previous_offset = 0.0
self.threshold = 12.5
def detect(
self, offset: float, timestamp_delta_ms: float, num_of_deltas: int, now_ms: int
) -> BandwidthUsage:
if num_of_deltas < 2:
return BandwidthUsage.NORMAL
T = min(num_of_deltas, MIN_NUM_DELTAS) * offset
if T > self.threshold:
if self.overuse_time is None:
self.overuse_time = timestamp_delta_ms / 2
else:
self.overuse_time += timestamp_delta_ms
self.overuse_counter += 1
if (
self.overuse_time > self.overuse_time_threshold
and self.overuse_counter > 1
and offset >= self.previous_offset
):
self.overuse_counter = 0
self.overuse_time = 0
self.hypothesis = BandwidthUsage.OVERUSING
elif T < -self.threshold:
self.overuse_counter = 0
self.overuse_time = None
self.hypothesis = BandwidthUsage.UNDERUSING
else:
self.overuse_counter = 0
self.overuse_time = None
self.hypothesis = BandwidthUsage.NORMAL
self.previous_offset = offset
self.update_threshold(T, now_ms)
return self.hypothesis
def state(self) -> BandwidthUsage:
return self.hypothesis
def update_threshold(self, modified_offset: float, now_ms: int) -> None:
if self.last_update_ms is None:
self.last_update_ms = now_ms
if abs(modified_offset) > self.threshold + MAX_ADAPT_OFFSET_MS:
self.last_update_ms = now_ms
return
k = self.k_down if abs(modified_offset) < self.threshold else self.k_up
time_delta_ms = min(now_ms - self.last_update_ms, 100)
self.threshold += k * (abs(modified_offset) - self.threshold) * time_delta_ms
self.threshold = max(6, min(self.threshold, 600))
self.last_update_ms = now_ms
class OveruseEstimator:
"""
Bandwidth overuse estimator.
Adapted from the webrtc.org codebase.
"""
def __init__(self) -> None:
self.E = [[100.0, 0.0], [0.0, 0.1]]
self._num_of_deltas = 0
self._offset = 0.0
self.previous_offset = 0.0
self.slope = 1 / 64
self.ts_delta_hist: List[float] = []
self.avg_noise = 0.0
self.var_noise = 50.0
self.process_noise = [1e-13, 1e-3]
def num_of_deltas(self) -> int:
return self._num_of_deltas
def offset(self) -> float:
return self._offset
def update(
self,
time_delta_ms: int,
timestamp_delta_ms: float,
size_delta: int,
current_hypothesis: BandwidthUsage,
now_ms: int,
):
min_frame_period = self.update_min_frame_period(timestamp_delta_ms)
t_ts_delta = time_delta_ms - timestamp_delta_ms
fs_delta = size_delta
self._num_of_deltas = min(self._num_of_deltas + 1, DELTA_COUNTER_MAX)
# update Kalman filter
self.E[0][0] += self.process_noise[0]
self.E[1][1] += self.process_noise[1]
if (
current_hypothesis == BandwidthUsage.OVERUSING
and self._offset < self.previous_offset
) or (
current_hypothesis == BandwidthUsage.UNDERUSING
and self._offset > self.previous_offset
):
self.E[1][1] += 10 * self.process_noise[1]
h = [fs_delta, 1.0]
Eh = [
self.E[0][0] * h[0] + self.E[0][1] * h[1],
self.E[1][0] * h[0] + self.E[1][1] * h[1],
]
# update noise estimate
residual = t_ts_delta - self.slope * h[0] - self._offset
if current_hypothesis == BandwidthUsage.NORMAL:
max_residual = 3.0 * math.sqrt(self.var_noise)
if abs(residual) < max_residual:
self.update_noise_estimate(residual, min_frame_period)
else:
self.update_noise_estimate(
-max_residual if residual < 0 else max_residual, min_frame_period
)
denom = self.var_noise + h[0] * Eh[0] + h[1] * Eh[1]
K = [Eh[0] / denom, Eh[1] / denom]
IKh = [[1.0 - K[0] * h[0], -K[0] * h[1]], [-K[1] * h[0], 1.0 - K[1] * h[1]]]
e00 = self.E[0][0]
e01 = self.E[0][1]
# update state
self.E[0][0] = e00 * IKh[0][0] + self.E[1][0] * IKh[0][1]
self.E[0][1] = e01 * IKh[0][0] + self.E[1][1] * IKh[0][1]
self.E[1][0] = e00 * IKh[1][0] + self.E[1][0] * IKh[1][1]
self.E[1][1] = e01 * IKh[1][0] + self.E[1][1] * IKh[1][1]
self.previous_offset = self._offset
self.slope += K[0] * residual
self._offset += K[1] * residual
def update_min_frame_period(self, ts_delta: float) -> float:
min_frame_period = ts_delta
if len(self.ts_delta_hist) >= MIN_FRAME_PERIOD_HISTORY_LENGTH:
self.ts_delta_hist.pop(0)
for old_ts_delta in self.ts_delta_hist:
min_frame_period = min(old_ts_delta, min_frame_period)
self.ts_delta_hist.append(ts_delta)
return min_frame_period
def update_noise_estimate(self, residual: float, ts_delta: float) -> None:
alpha = 0.01
if self._num_of_deltas > 10 * 30:
alpha = 0.002
beta = pow(1 - alpha, ts_delta * 30.0 / 1000.0)
self.avg_noise = beta * self.avg_noise + (1 - beta) * residual
self.var_noise = (
beta * self.var_noise + (1 - beta) * (self.avg_noise - residual) ** 2
)
if self.var_noise < 1:
self.var_noise = 1
class RateBucket:
def __init__(self, count: int = 0, value: int = 0) -> None:
self.count = count
self.value = value
def __eq__(self, other) -> bool:
return self.count == other.count and self.value == other.value
class RateCounter:
"""
Rate counter, which stores the amount received in 1ms buckets.
"""
def __init__(self, window_size: int, scale: int = 8000) -> None:
self._origin_index = 0
self._origin_ms: Optional[int] = None
self._scale = scale
self._window_size = window_size
self.reset()
def add(self, value: int, now_ms: int) -> None:
if self._origin_ms is None:
self._origin_ms = now_ms
else:
self._erase_old(now_ms)
index = (self._origin_index + now_ms - self._origin_ms) % self._window_size
self._buckets[index].count += 1
self._buckets[index].value += value
self._total.count += 1
self._total.value += value
def rate(self, now_ms: int) -> Optional[int]:
if self._origin_ms is not None:
self._erase_old(now_ms)
active_window_size = now_ms - self._origin_ms + 1
if self._total.count > 0 and active_window_size > 1:
return round(self._scale * self._total.value / active_window_size)
return None
def reset(self) -> None:
self._buckets = [RateBucket() for i in range(self._window_size)]
self._origin_index = 0
self._origin_ms = None
self._total = RateBucket()
def _erase_old(self, now_ms: int) -> None:
new_origin_ms = now_ms - self._window_size + 1
while self._origin_ms < new_origin_ms:
bucket = self._buckets[self._origin_index]
self._total.count -= bucket.count
self._total.value -= bucket.value
bucket.count = 0
bucket.value = 0
self._origin_index = (self._origin_index + 1) % self._window_size
self._origin_ms += 1
class RemoteBitrateEstimator:
def __init__(self) -> None:
self.incoming_bitrate = RateCounter(1000, 8000)
self.incoming_bitrate_initialized = True
self.inter_arrival = InterArrival(
(TIMESTAMP_GROUP_LENGTH_MS << INTER_ARRIVAL_SHIFT) // 1000, TIMESTAMP_TO_MS
)
self.estimator = OveruseEstimator()
self.detector = OveruseDetector()
self.rate_control = AimdRateControl()
self.last_update_ms: Optional[int] = None
self.ssrcs: Dict[int, int] = {}
def add(
self, arrival_time_ms: int, abs_send_time: int, payload_size: int, ssrc: int
) -> Optional[Tuple[int, List[int]]]:
timestamp = abs_send_time << 8
update_estimate = False
# make note of SSRC
self.ssrcs[ssrc] = arrival_time_ms
# update incoming bitrate
if self.incoming_bitrate.rate(arrival_time_ms) is not None:
self.incoming_bitrate_initialized = True
elif self.incoming_bitrate_initialized:
self.incoming_bitrate.reset()
self.incoming_bitrate_initialized = False
self.incoming_bitrate.add(payload_size, arrival_time_ms)
# calculate inter-arrival deltas
deltas = self.inter_arrival.compute_deltas(
timestamp, arrival_time_ms, payload_size
)
if deltas is not None:
timestamp_delta_ms = deltas.timestamp * TIMESTAMP_TO_MS
self.estimator.update(
deltas.arrival_time,
timestamp_delta_ms,
deltas.size,
self.detector.state(),
arrival_time_ms,
)
self.detector.detect(
self.estimator.offset(),
timestamp_delta_ms,
self.estimator.num_of_deltas(),
arrival_time_ms,
)
if not update_estimate:
if (
self.last_update_ms is None
or (arrival_time_ms - self.last_update_ms)
> self.rate_control.feedback_interval()
):
update_estimate = True
elif self.detector.state() == BandwidthUsage.OVERUSING:
update_estimate = True
if update_estimate:
target_bitrate = self.rate_control.update(
self.detector.state(),
self.incoming_bitrate.rate(arrival_time_ms),
arrival_time_ms,
)
if target_bitrate is not None:
self.last_update_ms = arrival_time_ms
return target_bitrate, list(self.ssrcs.keys())
return None
|