Lab046A: validate continuous soft-symbol frame scanning
This commit is contained in:
LittleSam129
@@ -719,3 +719,33 @@ Software gate: quick_gate PASS (438 тестов ядра, 6 функциона
|
||||
## Следующий этап
|
||||
|
||||
По решению ChatGPT после анализа Lab046: либо оптимизация накладных расходов серии кадров (преамбула/guard на кадр доминируют в goodput), либо подключение erasure-кода 12+3 для восстановления потери целых фрагментов поверх уже готового bit-level FEC, либо аппаратный прогон многопакетной передачи. RF, QPSK, JPEG и видео в этой сессии не запускались.
|
||||
|
||||
# Запись 026
|
||||
|
||||
## Дата
|
||||
|
||||
1 сентября 2026 года
|
||||
|
||||
## Тема
|
||||
|
||||
Lab046A: continuous multi-frame receive для 16 KiB объекта в одном soft-stream - только программно, RF TX = 0, аппаратная проверка не проводилась.
|
||||
|
||||
## Выполнено
|
||||
|
||||
Статус: **CONTINUOUS STREAM VALIDATION = PASS (NORMALIZED SOFT-SYMBOL LEVEL). CONTINUOUS SAMPLE/IQ VALIDATION = NOT PERFORMED. RF TX = 0. HARDWARE VALIDATION = NOT PERFORMED.** HackRF, hackrf_transfer, RF RX/TX, QPSK, JPEG и видео не запускались.
|
||||
|
||||
Проведён аудит Lab043-Lab046. Установлено, что предыдущий Lab046 уже собирал 16 KiB объект из 68 Lab045 кадров, но делал это по отдельным кадрам из TX-плана и не доказывал поиск нескольких кадров в одном continuous receive stream. Добавлен Lab046A слой в `experiments/lab046_bpsk_object_transport.py`: приёмник получает только один `float32` normalized soft-symbol stream, самостоятельно ищет PRBS7 candidates, fail-closed декодирует защищённый PHY header, вычисляет длину coded body только из принятого header, запускает существующие Lab045 FEC/CRC и передаёт fragment в `ObjectAssembler`.
|
||||
|
||||
TX offsets, список длин кадров, ожидаемое число 68, порядок/индексы TX, исходный объект, TX packets и coded bits scanner-у не передаются. TX-side `frame_starts` остаётся только в генераторе stream и используется тестами только для постфактум-сравнения после независимого receive. Scanner fail-closed: rejected candidate не останавливает поиск; cursor монотонен; `MAX_FRAMES` и `MAX_SEARCH_ITERATIONS` дают явную причину остановки.
|
||||
|
||||
Verifier read-only нашёл реальный дефект первого варианта: глобальный non-maximum suppression мог удалить настоящий старт кадра, если перед ним был более сильный ложный PRBS7 peak. Исправление: список candidates теперь содержит все позиции выше порога в порядке stream; повторный decode уже принятого кадра предотвращается только монотонным cursor после accepted frame. Добавлен регрессионный тест `test_stronger_false_peak_does_not_delete_overlapping_later_frame`.
|
||||
|
||||
Фактические continuous-сценарии для детерминированного объекта 16384 байта, SHA-256 `748013b5f670c82bd52650be5454555355dbd8528538753df4558f1c3408d515`: ideal continuous - candidates 68, valid headers 68, FEC PASS 68, CRC PASS 68/68, fragments 68, complete=True, SHA совпал, 0 байт-ошибок; background до/между/после 37/37/43 soft-символа - valid frames 68, SHA совпал; noisy continuous sigma=0.40 seed=46001 - pre-FEC 1791, post-FEC 0, CRC 68/0, complete=True, SHA совпал; malformed preamble + bad protected header - candidates 71, rejected 3, последующие valid frames 68, complete=True, SHA совпал.
|
||||
|
||||
Учёт: 68 фрагментов, 389 232 soft-symbols, 156 835 200 эквивалентных waveform samples, baseline airtime 19,604400 с при 8 MS/s, useful goodput 6,685846 кбит/с. Основной soft stream занимает 1 556 928 байт; correlation scores 1 551 812 байт; cumulative-energy 3 113 864 байта. Полный continuous `complex64` IQ stream не материализуется; при 156 835 200 samples он занял бы около 1,17 GiB до временных копий.
|
||||
|
||||
Документация обновлена в `docs/lab046_current_state.md`. Прямой запуск `python experiments/lab046_bpsk_object_transport.py --object-size 300 --sigma 0.4` проверен и работает. Software gate после исправлений: `python tools/quick_gate.py` PASS, `python -m pytest tests/ -q` 454 passed, 0 failed, 0 errors, 0 skipped, `git diff --check` чист. Git add/commit/fetch/pull/push не выполнялись.
|
||||
|
||||
## Следующий этап
|
||||
|
||||
Сначала ChatGPT смотрит Lab046A software result. После отдельного явного разрешения можно выбирать следующий шаг: аппаратная проверка многокадровой передачи, оптимизация накладных расходов серии кадров или erasure-FEC для потери целых фрагментов. RF TX, QPSK, JPEG и video автоматически не запускать.
|
||||
|
||||
@@ -2,7 +2,10 @@
|
||||
|
||||
## Статус
|
||||
|
||||
**SOFTWARE VALIDATED. RF TX = 0.
|
||||
**LAB046 SOFTWARE OBJECT TRANSPORT = COMPLETE.**
|
||||
|
||||
**CONTINUOUS STREAM VALIDATION = PASS (NORMALIZED SOFT-SYMBOL LEVEL). RF TX = 0.
|
||||
CONTINUOUS SAMPLE/IQ VALIDATION = NOT PERFORMED.
|
||||
HARDWARE VALIDATION = NOT PERFORMED.**
|
||||
|
||||
Lab046 — только программная лаба. HackRF и `hackrf_transfer` не
|
||||
@@ -46,6 +49,77 @@ fragment_count u16, original_size u32).
|
||||
использован и НЕ изменён — он остаётся будущим расширением для потери целых
|
||||
фрагментов, а не для битовых ошибок (это работа свёрточного кода).
|
||||
|
||||
После этого Lab046A закрыла оставшийся пробел: до неё Lab046 передавала каждый
|
||||
кадр отдельным вызовом и выбирала его из TX-плана по известному индексу. Функции
|
||||
поиска нескольких кадров в одном потоке не было. Теперь
|
||||
`scan_continuous_soft_stream` получает только один массив soft-символов,
|
||||
самостоятельно ищет PRBS7 после монотонного cursor, проверяет защищённый PHY
|
||||
header, вычисляет длину coded body из принятого packet length, выполняет готовые
|
||||
FEC/CRC Lab045 и передаёт принятый fragment в `ObjectAssembler`.
|
||||
|
||||
TX offsets, список длин кадров, ожидаемое число 68, порядок/индексы TX,
|
||||
исходный объект, TX packets и coded bits scanner-у не передаются. TX-side
|
||||
`frame_starts` остаётся только в генераторе потока и используется тестом для
|
||||
сравнения уже после независимого receive.
|
||||
|
||||
## Граница доказательства Lab046A
|
||||
|
||||
Lab046A работает на уровне **одного непрерывного нормализованного soft BPSK
|
||||
stream** (`float32`). Это разрешённый интерфейс после уже проверенного в Lab045
|
||||
sample-level front-end: PRBS7 detection, matched filtering, timing, CFO и
|
||||
phase/polarity recovery. Внутри нового длинного теста эти sample-level операции
|
||||
заново не моделируются для каждого кадра. Поэтому результат доказывает
|
||||
самостоятельный поиск границ, header-driven длину, FEC/CRC и сборку объекта в
|
||||
непрерывном soft stream, но не является аппаратной или непрерывной IQ-проверкой.
|
||||
|
||||
Каждый фрагмент содержит полный активный Lab045 PHY frame: 1280 бит PRBS7,
|
||||
144 бита защищённого header и 4300 coded bits. Точная полная sample-level форма
|
||||
с guard и хвостом ФРК учитывается как 2 306 400 отсчётов на кадр, однако все
|
||||
156 835 200 IQ-отсчётов одновременно в памяти не создаются. Это намеренное
|
||||
ограничение памяти, а не использование известных границ приёмником.
|
||||
|
||||
Scanner fail-closed:
|
||||
|
||||
- cursor всегда растёт, после принятого кадра переносится за вычисленный конец;
|
||||
- все PRBS7 candidates выше порога рассматриваются в порядке stream; после
|
||||
принятого кадра cursor переносится за вычисленный конец и не даёт
|
||||
декодировать этот кадр повторно;
|
||||
- плохой header, усечённое body, ошибка FEC/CRC отвергают candidate, но поиск
|
||||
следующего продолжается;
|
||||
- `MAX_FRAMES` и `MAX_SEARCH_ITERATIONS` дают явную причину остановки, а не
|
||||
частичный объект, объявленный полным;
|
||||
- scan продолжается до конца capture и не останавливается на числе 68.
|
||||
|
||||
## Сценарии continuous stream
|
||||
|
||||
| Сценарий | Получено |
|
||||
|---|---|
|
||||
| Идеальный, кадры подряд | candidates 68, valid headers 68, FEC PASS 68, CRC PASS 68/68, fragments 68, complete=True, SHA совпал, 0 байт-ошибок |
|
||||
| Фон до/между/после | 37/37/43 soft-символа, valid frames 68, complete=True, SHA совпал |
|
||||
| Непрерывный AWGN, sigma=0.40, seed=46001 | frames 68, pre-FEC 1791, post-FEC 0, CRC 68/0, complete=True, SHA совпал |
|
||||
| Valid-looking preamble + bad protected header | candidates 71, rejected 3, последующие valid frames 68, complete=True, SHA совпал |
|
||||
|
||||
Фон между кадрами добавлен только после кадров 7, 23 и 51: 11, 19 и 7
|
||||
soft-символов, всего 37. В baseline дополнительного inter-frame gap нет.
|
||||
Отдельный тест с валидным header для 12-байтного packet и повреждённым body
|
||||
проверяет, что scanner берёт длину именно из принятого header, отвергает этот
|
||||
candidate и затем находит настоящий кадр.
|
||||
|
||||
Verifier нашёл опасный вариант прежнего глобального non-maximum suppression:
|
||||
более сильный ложный PRBS7 peak мог удалить более поздний настоящий старт
|
||||
кадра. Поэтому suppression убран из списка кандидатов, а добавленный
|
||||
регрессионный тест проверяет, что ложный peak не скрывает настоящий кадр.
|
||||
|
||||
## Память continuous stream
|
||||
|
||||
Основной 68-кадровый soft stream содержит 389 232 значения `float32` и занимает
|
||||
1 556 928 байт (1,485 MiB). Массив correlation scores занимает 1 551 812 байт,
|
||||
накопленная энергия `float64` — 3 113 864 байта. FFT и denominator создают
|
||||
временное рабочее пространство, размер которого зависит от SciPy, но стратегия
|
||||
остается в пределах десятков MiB, а не нескольких GiB. Для сравнения, один
|
||||
материализованный continuous `complex64` IQ stream из 156 835 200 отсчётов
|
||||
занял бы около 1,17 GiB ещё до временных копий.
|
||||
|
||||
## Детерминированный объект
|
||||
|
||||
- Размер **16384 байта (16 KiB)**. Это не JPEG, а воспроизводимый
|
||||
@@ -118,8 +192,12 @@ count/original_size, невозможный индекс, конфликтующ
|
||||
| Число BPSK-кадров | 68 |
|
||||
| Всего информационных BPSK-символов | 389 232 (5724 × 68) |
|
||||
| Всего отсчётов формы сигнала | 156 835 200 (2 306 400 × 68) |
|
||||
| Оценка эфирного времени | 19,60 с |
|
||||
| Полезный goodput | 835,7 байт/с = 6,69 кбит/с |
|
||||
| Baseline: дополнительный inter-frame gap | 0 отсчётов |
|
||||
| Baseline: полная длительность | 19,604400 с |
|
||||
| Baseline: полезный goodput | 835,7 байт/с = 6,685846 кбит/с |
|
||||
| Фоновый тест: leading/inter-frame/trailing | 14 800 / 14 800 / 17 200 отсчётов |
|
||||
| Фоновый тест: полная длина и длительность | 156 882 000 отсчётов, 19,610250 с |
|
||||
| Фоновый тест: фактический goodput | 6,683852 кбит/с |
|
||||
| Полезных бит на символ | ≈ 0,337 |
|
||||
|
||||
Учёт кадра совпадает с Lab045: 2144 информационных бита, 6 tail-бит, 4300
|
||||
@@ -149,15 +227,17 @@ Goodput 6,69 кбит/с — это честная baseline при 20 ксимв
|
||||
проходят (фрагмент туда-обратно, короткий последний фрагмент, чужой
|
||||
object_id, невозможный индекс, несогласованный счётчик, усечённые и абсурдные
|
||||
метаданные без крупной аллокации, полный тракт PHY для сценариев A–F,
|
||||
отсутствие оракула, детерминизм и SHA объекта).
|
||||
отсутствие оракула, последовательный continuous scan, rejected candidate,
|
||||
MAX_FRAMES/MAX_SEARCH_ITERATIONS, регрессия против удаления настоящего кадра
|
||||
ложным PRBS7 peak, детерминизм и SHA объекта).
|
||||
- Полный тракт `объект → фрагменты → кадр Lab045 → канал → приём → сборка`
|
||||
проверен для всех шести сценариев значениями выше.
|
||||
|
||||
## Software gate
|
||||
|
||||
- `python tools/quick_gate.py`: PASS; 438 тестов ядра, 6 функциональных
|
||||
- `python tools/quick_gate.py`: PASS; 454 теста ядра, 6 функциональных
|
||||
проверок лабораторий выполнено, 13 штатно пропущено.
|
||||
- `python -m pytest tests/ -q`: 438 passed, 0 failed, 0 errors, 0 skipped.
|
||||
- `python -m pytest tests/ -q`: 454 passed, 0 failed, 0 errors, 0 skipped.
|
||||
Единственное предупреждение связано с отсутствием прав на создание
|
||||
`.pytest_cache`; на выполнение и результат тестов оно не повлияло.
|
||||
- `git diff --check`: чист.
|
||||
|
||||
@@ -32,11 +32,16 @@ from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
import hashlib
|
||||
import sys
|
||||
from dataclasses import dataclass, replace
|
||||
from pathlib import Path
|
||||
from typing import Callable
|
||||
|
||||
import numpy as np
|
||||
|
||||
if __package__ in (None, ""):
|
||||
sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
|
||||
|
||||
from experiments import lab045_hackrf_bpsk_fec as lab045
|
||||
from protocol import image_fragments as fragments
|
||||
|
||||
@@ -49,6 +54,10 @@ DEFAULT_OBJECT_SIZE = 16384 # 16 KiB, ровный размер, не JPEG
|
||||
DEFAULT_OBJECT_ID = 0x4C303436 # "L046"
|
||||
DEFAULT_OBJECT_SEED = 46
|
||||
|
||||
CONTINUOUS_PREAMBLE_THRESHOLD = 0.80
|
||||
DEFAULT_MAX_FRAMES = 256
|
||||
DEFAULT_MAX_SEARCH_ITERATIONS = 1024
|
||||
|
||||
|
||||
# ============================================================ Детерминированный объект
|
||||
|
||||
@@ -421,6 +430,453 @@ def run_object_transport(
|
||||
)
|
||||
|
||||
|
||||
# ============================================================ Lab046A: один continuous soft stream
|
||||
|
||||
|
||||
def _bits_to_float32_soft(bits: np.ndarray) -> np.ndarray:
|
||||
"""Преобразовать BPSK-биты в компактный нормализованный soft stream."""
|
||||
|
||||
values = np.asarray(bits)
|
||||
if values.ndim != 1 or not np.all((values == 0) | (values == 1)):
|
||||
raise ValueError("bits должны быть одномерной двоичной последовательностью")
|
||||
soft = values.astype(np.float32)
|
||||
soft *= np.float32(2.0)
|
||||
soft -= np.float32(1.0)
|
||||
return soft
|
||||
|
||||
|
||||
def malformed_header_candidate_soft() -> np.ndarray:
|
||||
"""Преамбула PRBS7 и намеренно неверная контрольная сумма PHY header."""
|
||||
|
||||
header = lab045.encode_phy_header(lab045.MIN_PACKET_BYTES).copy()
|
||||
checksum_bit = lab045.PHY_HEADER_FIELDS_BITS - 1
|
||||
first_copy = checksum_bit * lab045.HEADER_REPETITION
|
||||
header[first_copy:first_copy + 2] ^= 1
|
||||
candidate_bits = np.concatenate((lab045.lab044.preamble_bits(), header))
|
||||
return _bits_to_float32_soft(candidate_bits)
|
||||
|
||||
|
||||
@dataclass(frozen=True)
|
||||
class ContinuousSoftStreamPlan:
|
||||
"""TX-side описание потока; ``frame_starts`` запрещено передавать scanner-у."""
|
||||
|
||||
soft_symbols: np.ndarray
|
||||
frame_plans: tuple[FramePlan, ...]
|
||||
frame_starts: tuple[int, ...]
|
||||
leading_background_symbols: int
|
||||
inter_frame_background_symbols: int
|
||||
trailing_background_symbols: int
|
||||
malformed_candidate_symbols: int
|
||||
noise_sigma: float
|
||||
|
||||
@property
|
||||
def total_frame_symbols(self) -> int:
|
||||
return sum(frame.plan.frame_symbol_count for frame in self.frame_plans)
|
||||
|
||||
@property
|
||||
def frame_waveform_samples(self) -> int:
|
||||
return sum(frame.plan.waveform_sample_count for frame in self.frame_plans)
|
||||
|
||||
@property
|
||||
def extra_background_samples(self) -> int:
|
||||
symbols = (
|
||||
self.leading_background_symbols
|
||||
+ self.inter_frame_background_symbols
|
||||
+ self.trailing_background_symbols
|
||||
)
|
||||
return symbols * lab045.SAMPLES_PER_SYMBOL
|
||||
|
||||
@property
|
||||
def equivalent_total_samples(self) -> int:
|
||||
"""Полные Lab045-формы плюс намеренно добавленные фоновые области."""
|
||||
|
||||
auxiliary = self.malformed_candidate_symbols * lab045.SAMPLES_PER_SYMBOL
|
||||
return self.frame_waveform_samples + self.extra_background_samples + auxiliary
|
||||
|
||||
@property
|
||||
def equivalent_duration_seconds(self) -> float:
|
||||
return self.equivalent_total_samples / lab045.SAMPLE_RATE_HZ
|
||||
|
||||
|
||||
def build_continuous_soft_stream(
|
||||
frame_plans: list[FramePlan] | tuple[FramePlan, ...],
|
||||
*,
|
||||
leading_background_symbols: int = 0,
|
||||
inter_frame_background: dict[int, int] | None = None,
|
||||
trailing_background_symbols: int = 0,
|
||||
background_sigma: float = 0.12,
|
||||
noise_sigma: float = 0.0,
|
||||
seed: int = 46001,
|
||||
prepend_malformed_candidate: bool = False,
|
||||
) -> ContinuousSoftStreamPlan:
|
||||
"""Собрать один float32 stream без передачи RX известных границ кадров.
|
||||
|
||||
В stream представлены нормализованные soft BPSK-символы на выходе уже
|
||||
проверенного Lab045 timing/CFO/phase front-end. Полные sample-level формы
|
||||
Lab045 не материализуются: их точная длина учитывается свойствами плана.
|
||||
"""
|
||||
|
||||
frames = tuple(frame_plans)
|
||||
if not frames:
|
||||
raise ValueError("нужен хотя бы один кадр")
|
||||
if leading_background_symbols < 0 or trailing_background_symbols < 0:
|
||||
raise ValueError("длина фоновой области не может быть отрицательной")
|
||||
if background_sigma < 0.0 or noise_sigma < 0.0:
|
||||
raise ValueError("sigma не может быть отрицательной")
|
||||
gaps = dict(inter_frame_background or {})
|
||||
for frame_index, symbol_count in gaps.items():
|
||||
if not 0 <= frame_index < len(frames) - 1 or symbol_count < 0:
|
||||
raise ValueError("некорректная inter-frame область")
|
||||
|
||||
random_generator = np.random.default_rng(seed)
|
||||
chunks: list[np.ndarray] = []
|
||||
frame_starts: list[int] = []
|
||||
current_offset = 0
|
||||
|
||||
def append_background(symbol_count: int) -> None:
|
||||
nonlocal current_offset
|
||||
if symbol_count == 0:
|
||||
return
|
||||
background = random_generator.normal(
|
||||
0.0,
|
||||
background_sigma,
|
||||
size=symbol_count,
|
||||
).astype(np.float32)
|
||||
chunks.append(background)
|
||||
current_offset += symbol_count
|
||||
|
||||
append_background(leading_background_symbols)
|
||||
malformed_symbols = 0
|
||||
if prepend_malformed_candidate:
|
||||
malformed = malformed_header_candidate_soft()
|
||||
chunks.append(malformed)
|
||||
malformed_symbols = int(malformed.size)
|
||||
current_offset += malformed_symbols
|
||||
|
||||
inter_frame_symbols = 0
|
||||
for frame_index, frame in enumerate(frames):
|
||||
frame_starts.append(current_offset)
|
||||
frame_soft = _bits_to_float32_soft(frame.plan.frame_bits)
|
||||
chunks.append(frame_soft)
|
||||
current_offset += int(frame_soft.size)
|
||||
if frame_index < len(frames) - 1:
|
||||
gap = int(gaps.get(frame_index, 0))
|
||||
inter_frame_symbols += gap
|
||||
append_background(gap)
|
||||
append_background(trailing_background_symbols)
|
||||
|
||||
stream = np.concatenate(chunks, dtype=np.float32)
|
||||
if noise_sigma > 0.0:
|
||||
noise = random_generator.normal(0.0, noise_sigma, size=stream.size)
|
||||
stream += noise.astype(np.float32)
|
||||
return ContinuousSoftStreamPlan(
|
||||
soft_symbols=stream,
|
||||
frame_plans=frames,
|
||||
frame_starts=tuple(frame_starts),
|
||||
leading_background_symbols=leading_background_symbols,
|
||||
inter_frame_background_symbols=inter_frame_symbols,
|
||||
trailing_background_symbols=trailing_background_symbols,
|
||||
malformed_candidate_symbols=malformed_symbols,
|
||||
noise_sigma=noise_sigma,
|
||||
)
|
||||
|
||||
|
||||
def _normalized_soft_preamble_correlation(soft_stream: np.ndarray) -> np.ndarray:
|
||||
"""Нормированная корреляция PRBS7 через проверенный helper Lab043."""
|
||||
|
||||
preamble = _bits_to_float32_soft(lab045.lab044.preamble_bits())
|
||||
if soft_stream.size < preamble.size:
|
||||
return np.empty(0, dtype=np.float32)
|
||||
scores = np.asarray(
|
||||
lab045.lab044.lab043._normalized_sliding_correlation(soft_stream, preamble),
|
||||
dtype=np.float32,
|
||||
)
|
||||
scores[~np.isfinite(scores)] = 0.0
|
||||
return np.clip(scores, 0.0, 1.0)
|
||||
|
||||
|
||||
def _preamble_candidate_indexes(scores: np.ndarray, threshold: float) -> np.ndarray:
|
||||
"""Вернуть все пороговые кандидаты в порядке потока, ничего не удаляя заранее."""
|
||||
|
||||
return np.flatnonzero(scores >= threshold)
|
||||
|
||||
|
||||
@dataclass(frozen=True)
|
||||
class ContinuousFrameReception:
|
||||
candidate_start: int
|
||||
frame_end: int
|
||||
preamble_score: float
|
||||
packet_bytes: int
|
||||
fragment_index: int | None
|
||||
reception: lab045.IndependentReception
|
||||
|
||||
|
||||
@dataclass(frozen=True)
|
||||
class ContinuousReceiveResult:
|
||||
stream_symbol_count: int
|
||||
preamble_candidates: int
|
||||
valid_phy_headers: int
|
||||
fec_pass_count: int
|
||||
crc_pass_count: int
|
||||
crc_fail_count: int
|
||||
valid_frame_count: int
|
||||
rejected_candidates: int
|
||||
valid_fragments: int
|
||||
duplicate_fragments: int
|
||||
conflicting_fragments: int
|
||||
rejected_fragments: int
|
||||
missing_indexes: tuple[int, ...]
|
||||
object_complete: bool
|
||||
recovered_object: bytes | None
|
||||
candidate_starts: tuple[int, ...]
|
||||
cursor_history: tuple[int, ...]
|
||||
search_iterations: int
|
||||
termination_reason: str
|
||||
max_frames_reached: bool
|
||||
max_search_iterations_reached: bool
|
||||
correlation_storage_bytes: int
|
||||
frame_receptions: tuple[ContinuousFrameReception, ...]
|
||||
|
||||
|
||||
def scan_continuous_soft_stream(
|
||||
soft_stream: np.ndarray,
|
||||
*,
|
||||
preamble_threshold: float = CONTINUOUS_PREAMBLE_THRESHOLD,
|
||||
max_frames: int = DEFAULT_MAX_FRAMES,
|
||||
max_search_iterations: int = DEFAULT_MAX_SEARCH_ITERATIONS,
|
||||
) -> ContinuousReceiveResult:
|
||||
"""Независимо найти и декодировать все кадры в одном soft stream.
|
||||
|
||||
Единственный вход с данными — сам stream. TX offsets, число кадров,
|
||||
fragment order, packets, coded bits и исходный объект функции неизвестны.
|
||||
"""
|
||||
|
||||
try:
|
||||
stream = np.asarray(soft_stream, dtype=np.float32)
|
||||
except (TypeError, ValueError):
|
||||
stream = np.empty((0, 0), dtype=np.float32)
|
||||
if not 0.0 < preamble_threshold <= 1.0:
|
||||
raise ValueError("preamble_threshold должен лежать в (0, 1]")
|
||||
if max_frames < 0 or max_search_iterations < 0:
|
||||
raise ValueError("safety limits не могут быть отрицательными")
|
||||
|
||||
assembler = ObjectAssembler()
|
||||
scores = np.empty(0, dtype=np.float32)
|
||||
if stream.ndim == 1 and stream.size and np.all(np.isfinite(stream)):
|
||||
scores = _normalized_soft_preamble_correlation(stream)
|
||||
candidate_indexes = _preamble_candidate_indexes(scores, preamble_threshold)
|
||||
|
||||
cursor = 0
|
||||
cursor_history = [cursor]
|
||||
candidate_starts: list[int] = []
|
||||
frame_receptions: list[ContinuousFrameReception] = []
|
||||
preamble_candidates = 0
|
||||
valid_headers = 0
|
||||
fec_pass = 0
|
||||
crc_pass = 0
|
||||
crc_fail = 0
|
||||
valid_frames = 0
|
||||
rejected_candidates = 0
|
||||
search_iterations = 0
|
||||
termination_reason = "END_OF_STREAM"
|
||||
max_frames_reached = False
|
||||
max_iterations_reached = False
|
||||
|
||||
while True:
|
||||
position = int(np.searchsorted(candidate_indexes, cursor, side="left"))
|
||||
if position >= candidate_indexes.size:
|
||||
break
|
||||
candidate_start = int(candidate_indexes[position])
|
||||
if search_iterations >= max_search_iterations:
|
||||
termination_reason = "MAX_SEARCH_ITERATIONS_REACHED"
|
||||
max_iterations_reached = True
|
||||
break
|
||||
if valid_frames >= max_frames:
|
||||
termination_reason = "MAX_FRAMES_REACHED"
|
||||
max_frames_reached = True
|
||||
break
|
||||
|
||||
search_iterations += 1
|
||||
preamble_candidates += 1
|
||||
candidate_starts.append(candidate_start)
|
||||
preamble_end = candidate_start + lab045.PREAMBLE_BIT_COUNT
|
||||
header_end = preamble_end + lab045.PHY_HEADER_BIT_COUNT
|
||||
header = lab045.decode_phy_header_soft(stream[preamble_end:header_end])
|
||||
if not header["passed"]:
|
||||
rejected_candidates += 1
|
||||
cursor = candidate_start + 1
|
||||
cursor_history.append(cursor)
|
||||
continue
|
||||
|
||||
valid_headers += 1
|
||||
packet_bytes = int(header["packet_bytes"])
|
||||
coded_length = lab045.conv.encoded_bit_count(packet_bytes * 8)
|
||||
frame_end = header_end + coded_length
|
||||
if frame_end > stream.size:
|
||||
rejected_candidates += 1
|
||||
cursor = candidate_start + 1
|
||||
cursor_history.append(cursor)
|
||||
continue
|
||||
|
||||
reception = lab045.analyze_post_preamble_soft(stream[preamble_end:frame_end])
|
||||
if reception.analysis["fec_decode_pass"]:
|
||||
fec_pass += 1
|
||||
fragment = None
|
||||
fragment_index = None
|
||||
if reception.analysis["crc"] == lab045.CRC_PASS and reception.payload_bytes is not None:
|
||||
crc_pass += 1
|
||||
valid_frames += 1
|
||||
fragment = recover_fragment_from_payload(reception.payload_bytes)
|
||||
if fragment is not None:
|
||||
fragment_index = int(fragment.fragment_index)
|
||||
assembler.offer(fragment)
|
||||
else:
|
||||
if reception.analysis["crc"] == lab045.CRC_FAIL:
|
||||
crc_fail += 1
|
||||
rejected_candidates += 1
|
||||
|
||||
frame_receptions.append(ContinuousFrameReception(
|
||||
candidate_start=candidate_start,
|
||||
frame_end=frame_end,
|
||||
preamble_score=float(scores[candidate_start]),
|
||||
packet_bytes=packet_bytes,
|
||||
fragment_index=fragment_index,
|
||||
reception=reception,
|
||||
))
|
||||
cursor = frame_end
|
||||
cursor_history.append(cursor)
|
||||
|
||||
recovered = assembler.recovered_object()
|
||||
if stream.ndim != 1 or not stream.size or not np.all(np.isfinite(stream)):
|
||||
termination_reason = "INVALID_STREAM"
|
||||
return ContinuousReceiveResult(
|
||||
stream_symbol_count=int(stream.size),
|
||||
preamble_candidates=preamble_candidates,
|
||||
valid_phy_headers=valid_headers,
|
||||
fec_pass_count=fec_pass,
|
||||
crc_pass_count=crc_pass,
|
||||
crc_fail_count=crc_fail,
|
||||
valid_frame_count=valid_frames,
|
||||
rejected_candidates=rejected_candidates,
|
||||
valid_fragments=assembler.counters.accepted,
|
||||
duplicate_fragments=assembler.counters.duplicates,
|
||||
conflicting_fragments=assembler.counters.conflicting,
|
||||
rejected_fragments=assembler.counters.rejected,
|
||||
missing_indexes=tuple(assembler.missing_indexes()),
|
||||
object_complete=assembler.object_complete(),
|
||||
recovered_object=recovered,
|
||||
candidate_starts=tuple(candidate_starts),
|
||||
cursor_history=tuple(cursor_history),
|
||||
search_iterations=search_iterations,
|
||||
termination_reason=termination_reason,
|
||||
max_frames_reached=max_frames_reached,
|
||||
max_search_iterations_reached=max_iterations_reached,
|
||||
correlation_storage_bytes=int(scores.nbytes),
|
||||
frame_receptions=tuple(frame_receptions),
|
||||
)
|
||||
|
||||
|
||||
@dataclass(frozen=True)
|
||||
class ContinuousValidationResult:
|
||||
stream_plan: ContinuousSoftStreamPlan
|
||||
receive: ContinuousReceiveResult
|
||||
object_sha256: str
|
||||
recovered_object_sha256: str | None
|
||||
object_byte_errors: int | None
|
||||
sha_match: bool
|
||||
pre_fec_coded_bit_errors: int
|
||||
post_fec_protocol_bit_errors: int
|
||||
|
||||
|
||||
def _continuous_post_receive_diagnostics(
|
||||
receive: ContinuousReceiveResult,
|
||||
stream_plan: ContinuousSoftStreamPlan,
|
||||
) -> tuple[int, int]:
|
||||
"""Сравнить с TX только после независимых detector/header/FEC/CRC решений."""
|
||||
|
||||
frame_by_index = {frame.fragment_index: frame for frame in stream_plan.frame_plans}
|
||||
pre_fec_errors = 0
|
||||
post_fec_errors = 0
|
||||
for record in receive.frame_receptions:
|
||||
if record.fragment_index is None:
|
||||
continue
|
||||
frame = frame_by_index.get(record.fragment_index)
|
||||
if frame is None:
|
||||
continue
|
||||
post_start = record.candidate_start + lab045.PREAMBLE_BIT_COUNT
|
||||
post_soft = stream_plan.soft_symbols[post_start:record.frame_end]
|
||||
diagnostic = lab045.diagnostic_pre_and_post_fec(
|
||||
record.reception,
|
||||
post_soft,
|
||||
frame.plan,
|
||||
)
|
||||
pre = diagnostic["pre_fec_coded"]["bit_errors"]
|
||||
post = diagnostic["post_fec_packet"]["bit_errors"]
|
||||
if pre is not None:
|
||||
pre_fec_errors += int(pre)
|
||||
if post is not None:
|
||||
post_fec_errors += int(post)
|
||||
return pre_fec_errors, post_fec_errors
|
||||
|
||||
|
||||
def run_continuous_object_transport(
|
||||
data: bytes,
|
||||
*,
|
||||
object_id: int = DEFAULT_OBJECT_ID,
|
||||
leading_background_symbols: int = 0,
|
||||
inter_frame_background: dict[int, int] | None = None,
|
||||
trailing_background_symbols: int = 0,
|
||||
noise_sigma: float = 0.0,
|
||||
seed: int = 46001,
|
||||
prepend_malformed_candidate: bool = False,
|
||||
max_frames: int = DEFAULT_MAX_FRAMES,
|
||||
max_search_iterations: int = DEFAULT_MAX_SEARCH_ITERATIONS,
|
||||
) -> ContinuousValidationResult:
|
||||
"""Построить TX stream, независимо принять его и лишь затем сравнить с TX."""
|
||||
|
||||
frame_plans = build_object_frames(data, object_id)
|
||||
stream_plan = build_continuous_soft_stream(
|
||||
frame_plans,
|
||||
leading_background_symbols=leading_background_symbols,
|
||||
inter_frame_background=inter_frame_background,
|
||||
trailing_background_symbols=trailing_background_symbols,
|
||||
noise_sigma=noise_sigma,
|
||||
seed=seed,
|
||||
prepend_malformed_candidate=prepend_malformed_candidate,
|
||||
)
|
||||
receive = scan_continuous_soft_stream(
|
||||
stream_plan.soft_symbols,
|
||||
max_frames=max_frames,
|
||||
max_search_iterations=max_search_iterations,
|
||||
)
|
||||
pre_fec_errors, post_fec_errors = _continuous_post_receive_diagnostics(
|
||||
receive,
|
||||
stream_plan,
|
||||
)
|
||||
recovered = receive.recovered_object
|
||||
recovered_sha = object_sha256(recovered) if recovered is not None else None
|
||||
original_sha = object_sha256(data)
|
||||
if recovered is None:
|
||||
byte_errors = None
|
||||
elif len(recovered) != len(data):
|
||||
byte_errors = max(len(recovered), len(data))
|
||||
else:
|
||||
byte_errors = int(np.count_nonzero(
|
||||
np.frombuffer(recovered, dtype=np.uint8)
|
||||
!= np.frombuffer(data, dtype=np.uint8)
|
||||
))
|
||||
return ContinuousValidationResult(
|
||||
stream_plan=stream_plan,
|
||||
receive=receive,
|
||||
object_sha256=original_sha,
|
||||
recovered_object_sha256=recovered_sha,
|
||||
object_byte_errors=byte_errors,
|
||||
sha_match=(recovered is not None and recovered_sha == original_sha),
|
||||
pre_fec_coded_bit_errors=pre_fec_errors,
|
||||
post_fec_protocol_bit_errors=post_fec_errors,
|
||||
)
|
||||
|
||||
|
||||
# ============================================================ Airtime / goodput
|
||||
|
||||
|
||||
@@ -451,6 +907,35 @@ def goodput_report(result: ObjectTransportResult) -> dict:
|
||||
}
|
||||
|
||||
|
||||
def continuous_accounting_report(result: ContinuousValidationResult) -> dict:
|
||||
"""Длины полной Lab045-формы и фактических тестовых областей Lab046A."""
|
||||
|
||||
plan = result.stream_plan
|
||||
baseline_duration = plan.frame_waveform_samples / lab045.SAMPLE_RATE_HZ
|
||||
actual_duration = plan.equivalent_duration_seconds
|
||||
useful_bits = len(result.receive.recovered_object or b"") * 8
|
||||
return {
|
||||
"total_frame_symbols": plan.total_frame_symbols,
|
||||
"frame_waveform_samples": plan.frame_plans[0].plan.waveform_sample_count,
|
||||
"all_frame_waveform_samples": plan.frame_waveform_samples,
|
||||
"leading_samples": plan.leading_background_symbols * lab045.SAMPLES_PER_SYMBOL,
|
||||
"inter_frame_gap_samples": plan.inter_frame_background_symbols * lab045.SAMPLES_PER_SYMBOL,
|
||||
"trailing_samples": plan.trailing_background_symbols * lab045.SAMPLES_PER_SYMBOL,
|
||||
"total_stream_samples": plan.equivalent_total_samples,
|
||||
"baseline_duration_seconds": baseline_duration,
|
||||
"actual_duration_seconds": actual_duration,
|
||||
"baseline_goodput_kbit_per_second": (
|
||||
useful_bits / baseline_duration / 1000.0 if baseline_duration else 0.0
|
||||
),
|
||||
"actual_goodput_kbit_per_second": (
|
||||
useful_bits / actual_duration / 1000.0 if actual_duration else 0.0
|
||||
),
|
||||
"soft_stream_bytes": int(plan.soft_symbols.nbytes),
|
||||
"correlation_scores_bytes": result.receive.correlation_storage_bytes,
|
||||
"cumulative_energy_bytes": int((plan.soft_symbols.size + 1) * 8),
|
||||
}
|
||||
|
||||
|
||||
# ============================================================ Функциональные тесты
|
||||
|
||||
|
||||
@@ -477,6 +962,12 @@ def run_functional_tests() -> None:
|
||||
recovered = recover_fragment_from_payload(padded)
|
||||
assert recovered is not None and recovered.data == last.data
|
||||
|
||||
# Короткая Lab046A-проверка: два кадра находятся в одном stream без offsets.
|
||||
continuous = run_continuous_object_transport(obj[:300])
|
||||
assert continuous.receive.valid_frame_count == 2
|
||||
assert continuous.receive.preamble_candidates == 2
|
||||
assert continuous.sha_match
|
||||
|
||||
|
||||
# ============================================================ main
|
||||
|
||||
@@ -495,6 +986,30 @@ def main() -> int:
|
||||
channel=make_noisy_channel(arguments.sigma, base_seed=46000),
|
||||
)
|
||||
report = goodput_report(ideal)
|
||||
background_gaps = {
|
||||
frame_index: symbol_count
|
||||
for frame_index, symbol_count in ((7, 11), (23, 19), (51, 7))
|
||||
if frame_index < ideal.fragment_count - 1
|
||||
}
|
||||
continuous_ideal = run_continuous_object_transport(obj)
|
||||
continuous_background = run_continuous_object_transport(
|
||||
obj,
|
||||
leading_background_symbols=37,
|
||||
inter_frame_background=background_gaps,
|
||||
trailing_background_symbols=43,
|
||||
seed=464,
|
||||
)
|
||||
continuous_noisy = run_continuous_object_transport(
|
||||
obj,
|
||||
noise_sigma=arguments.sigma,
|
||||
seed=46001,
|
||||
)
|
||||
continuous_malformed = run_continuous_object_transport(
|
||||
obj,
|
||||
prepend_malformed_candidate=True,
|
||||
)
|
||||
continuous_accounting = continuous_accounting_report(continuous_ideal)
|
||||
background_accounting = continuous_accounting_report(continuous_background)
|
||||
|
||||
print("Lab046 software object transport")
|
||||
print(f"object size = {ideal.object_size} bytes")
|
||||
@@ -515,6 +1030,32 @@ def main() -> int:
|
||||
print(f"post-FEC protocol errors = {noisy.post_fec_protocol_bit_errors}")
|
||||
print(f"crc pass/fail = {noisy.crc_pass_count}/{noisy.crc_fail_count}")
|
||||
print(f"object complete = {noisy.object_complete}, sha match = {noisy.sha_match}, byte errors = {noisy.object_byte_errors}")
|
||||
print("--- Lab046A ideal continuous soft stream ---")
|
||||
print(f"preamble candidates = {continuous_ideal.receive.preamble_candidates}")
|
||||
print(f"valid headers/frames = {continuous_ideal.receive.valid_phy_headers}/{continuous_ideal.receive.valid_frame_count}")
|
||||
print(f"fec/crc pass = {continuous_ideal.receive.fec_pass_count}/{continuous_ideal.receive.crc_pass_count}")
|
||||
print(f"rejected candidates = {continuous_ideal.receive.rejected_candidates}")
|
||||
print(f"object complete = {continuous_ideal.receive.object_complete}, sha match = {continuous_ideal.sha_match}, byte errors = {continuous_ideal.object_byte_errors}")
|
||||
print("--- Lab046A background continuous soft stream ---")
|
||||
print(f"leading/inter-frame/trailing symbols = {continuous_background.stream_plan.leading_background_symbols}/{continuous_background.stream_plan.inter_frame_background_symbols}/{continuous_background.stream_plan.trailing_background_symbols}")
|
||||
print(f"valid frames = {continuous_background.receive.valid_frame_count}, sha match = {continuous_background.sha_match}")
|
||||
print(f"actual equivalent samples/duration = {background_accounting['total_stream_samples']}/{background_accounting['actual_duration_seconds']:.6f} s")
|
||||
print(f"--- Lab046A noisy continuous soft stream (sigma={arguments.sigma}) ---")
|
||||
print(f"valid frames = {continuous_noisy.receive.valid_frame_count}")
|
||||
print(f"pre/post-FEC errors = {continuous_noisy.pre_fec_coded_bit_errors}/{continuous_noisy.post_fec_protocol_bit_errors}")
|
||||
print(f"crc pass/fail = {continuous_noisy.receive.crc_pass_count}/{continuous_noisy.receive.crc_fail_count}")
|
||||
print(f"object complete = {continuous_noisy.receive.object_complete}, sha match = {continuous_noisy.sha_match}")
|
||||
print("--- Lab046A malformed candidate ---")
|
||||
print(f"preamble candidates = {continuous_malformed.receive.preamble_candidates}")
|
||||
print(f"rejected candidates = {continuous_malformed.receive.rejected_candidates}")
|
||||
print(f"valid frames = {continuous_malformed.receive.valid_frame_count}, sha match = {continuous_malformed.sha_match}")
|
||||
print("--- Lab046A accounting ---")
|
||||
print(f"frame waveform samples = {continuous_accounting['frame_waveform_samples']}")
|
||||
print(f"all frame waveform samples = {continuous_accounting['all_frame_waveform_samples']}")
|
||||
print(f"inter-frame gap samples = {continuous_accounting['inter_frame_gap_samples']}")
|
||||
print(f"airtime = {continuous_accounting['baseline_duration_seconds']:.6f} s")
|
||||
print(f"useful goodput = {continuous_accounting['baseline_goodput_kbit_per_second']:.6f} kbit/s")
|
||||
print(f"soft/correlation/cumulative-energy bytes = {continuous_accounting['soft_stream_bytes']}/{continuous_accounting['correlation_scores_bytes']}/{continuous_accounting['cumulative_energy_bytes']}")
|
||||
print("RF TX = 0 (software-only)")
|
||||
return 0
|
||||
|
||||
|
||||
@@ -7,6 +7,7 @@
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import inspect
|
||||
import struct
|
||||
|
||||
import numpy as np
|
||||
@@ -243,3 +244,224 @@ def test_receiver_ignores_original_object_on_corruption() -> None:
|
||||
|
||||
def test_functional_self_check_runs() -> None:
|
||||
lab046.run_functional_tests()
|
||||
|
||||
|
||||
# ------------------------------------------------------------ Lab046A: continuous stream
|
||||
|
||||
|
||||
def _small_continuous_stream(
|
||||
*,
|
||||
leading: int = 0,
|
||||
gaps: dict[int, int] | None = None,
|
||||
trailing: int = 0,
|
||||
) -> tuple[bytes, lab046.ContinuousSoftStreamPlan]:
|
||||
obj = lab046.make_deterministic_object(300, seed=460)
|
||||
plan = lab046.build_continuous_soft_stream(
|
||||
lab046.build_object_frames(obj),
|
||||
leading_background_symbols=leading,
|
||||
inter_frame_background=gaps,
|
||||
trailing_background_symbols=trailing,
|
||||
seed=461,
|
||||
)
|
||||
return obj, plan
|
||||
|
||||
|
||||
def test_two_sequential_frames_found_without_known_offsets() -> None:
|
||||
obj, stream_plan = _small_continuous_stream()
|
||||
received = lab046.scan_continuous_soft_stream(stream_plan.soft_symbols)
|
||||
assert received.valid_frame_count == 2
|
||||
assert received.recovered_object == obj
|
||||
# Сравнение с TX offsets разрешено только здесь, после независимого receive.
|
||||
assert received.candidate_starts == stream_plan.frame_starts
|
||||
|
||||
|
||||
def test_many_sequential_frames_are_found() -> None:
|
||||
obj = lab046.make_deterministic_object(244 * 10, seed=462)
|
||||
result = lab046.run_continuous_object_transport(obj)
|
||||
assert result.receive.valid_frame_count == 10
|
||||
assert result.receive.preamble_candidates == 10
|
||||
assert result.sha_match
|
||||
|
||||
|
||||
def test_first_frame_need_not_start_at_zero() -> None:
|
||||
obj, stream_plan = _small_continuous_stream(leading=37)
|
||||
received = lab046.scan_continuous_soft_stream(stream_plan.soft_symbols)
|
||||
assert received.candidate_starts[0] == 37
|
||||
assert received.recovered_object == obj
|
||||
|
||||
|
||||
def test_trailing_noise_is_ignored() -> None:
|
||||
obj, stream_plan = _small_continuous_stream(trailing=43)
|
||||
received = lab046.scan_continuous_soft_stream(stream_plan.soft_symbols)
|
||||
assert received.valid_frame_count == 2
|
||||
assert received.preamble_candidates == 2
|
||||
assert received.recovered_object == obj
|
||||
|
||||
|
||||
def test_inter_frame_background_does_not_hide_next_frame() -> None:
|
||||
obj, stream_plan = _small_continuous_stream(gaps={0: 19})
|
||||
received = lab046.scan_continuous_soft_stream(stream_plan.soft_symbols)
|
||||
assert received.candidate_starts[1] - received.candidate_starts[0] == 5724 + 19
|
||||
assert received.recovered_object == obj
|
||||
|
||||
|
||||
def test_same_frame_is_not_decoded_twice() -> None:
|
||||
_, stream_plan = _small_continuous_stream()
|
||||
received = lab046.scan_continuous_soft_stream(stream_plan.soft_symbols)
|
||||
assert len(received.candidate_starts) == len(set(received.candidate_starts))
|
||||
assert received.preamble_candidates == received.valid_frame_count == 2
|
||||
|
||||
|
||||
def test_malformed_candidate_rejected_and_later_good_frame_found() -> None:
|
||||
obj = lab046.make_deterministic_object(300, seed=463)
|
||||
result = lab046.run_continuous_object_transport(
|
||||
obj,
|
||||
prepend_malformed_candidate=True,
|
||||
)
|
||||
assert result.receive.rejected_candidates > 0
|
||||
assert result.receive.valid_frame_count == 2
|
||||
assert result.receive.candidate_starts[0] == 0
|
||||
assert result.sha_match
|
||||
|
||||
|
||||
def test_stronger_false_peak_does_not_delete_overlapping_later_frame() -> None:
|
||||
obj = lab046.make_deterministic_object(100, seed=999)
|
||||
frame = lab046.build_object_frames(obj)[0]
|
||||
preamble = lab046._bits_to_float32_soft(lab046.lab045.lab044.preamble_bits())
|
||||
true_frame = lab046._bits_to_float32_soft(frame.plan.frame_bits)
|
||||
true_start = 508 # четыре периода PRBS7 после ложного старта
|
||||
stream = np.zeros(true_start + true_frame.size, dtype=np.float32)
|
||||
stream[:preamble.size] = preamble
|
||||
stream[true_start:true_start + true_frame.size] = true_frame
|
||||
stream[preamble.size:true_start + preamble.size] *= np.float32(0.7)
|
||||
|
||||
scores = lab046._normalized_soft_preamble_correlation(stream)
|
||||
assert scores[0] > scores[true_start]
|
||||
received = lab046.scan_continuous_soft_stream(stream)
|
||||
assert received.rejected_candidates > 0
|
||||
assert true_start in received.candidate_starts
|
||||
assert received.valid_frame_count == 1
|
||||
assert received.recovered_object == obj
|
||||
|
||||
|
||||
def test_validated_header_drives_candidate_body_length() -> None:
|
||||
obj, stream_plan = _small_continuous_stream()
|
||||
packet_bytes = lab046.lab045.MIN_PACKET_BYTES
|
||||
coded_symbols = lab046.lab045.conv.encoded_bit_count(packet_bytes * 8)
|
||||
prefix_bits = np.concatenate((
|
||||
lab046.lab045.lab044.preamble_bits(),
|
||||
lab046.lab045.encode_phy_header(packet_bytes),
|
||||
np.zeros(coded_symbols, dtype=np.uint8),
|
||||
))
|
||||
stream = np.concatenate((
|
||||
lab046._bits_to_float32_soft(prefix_bits),
|
||||
stream_plan.soft_symbols,
|
||||
))
|
||||
received = lab046.scan_continuous_soft_stream(stream)
|
||||
assert received.frame_receptions[0].packet_bytes == packet_bytes
|
||||
assert received.frame_receptions[0].frame_end == len(prefix_bits)
|
||||
assert received.rejected_candidates >= 1
|
||||
assert received.recovered_object == obj
|
||||
|
||||
|
||||
def test_cursor_is_strictly_monotonic() -> None:
|
||||
_, stream_plan = _small_continuous_stream(leading=7, gaps={0: 13}, trailing=5)
|
||||
received = lab046.scan_continuous_soft_stream(stream_plan.soft_symbols)
|
||||
assert all(
|
||||
later > earlier
|
||||
for earlier, later in zip(received.cursor_history, received.cursor_history[1:])
|
||||
)
|
||||
|
||||
|
||||
def test_max_search_iterations_is_fail_closed() -> None:
|
||||
_, stream_plan = _small_continuous_stream()
|
||||
received = lab046.scan_continuous_soft_stream(
|
||||
stream_plan.soft_symbols,
|
||||
max_search_iterations=1,
|
||||
)
|
||||
assert received.max_search_iterations_reached
|
||||
assert received.termination_reason == "MAX_SEARCH_ITERATIONS_REACHED"
|
||||
assert received.valid_frame_count == 1
|
||||
assert not received.object_complete
|
||||
|
||||
|
||||
def test_max_frames_is_fail_closed() -> None:
|
||||
_, stream_plan = _small_continuous_stream()
|
||||
received = lab046.scan_continuous_soft_stream(
|
||||
stream_plan.soft_symbols,
|
||||
max_frames=1,
|
||||
)
|
||||
assert received.max_frames_reached
|
||||
assert received.termination_reason == "MAX_FRAMES_REACHED"
|
||||
assert received.valid_frame_count == 1
|
||||
assert not received.object_complete
|
||||
|
||||
|
||||
def test_continuous_ideal_16_kib_object_is_exact() -> None:
|
||||
obj = lab046.make_deterministic_object()
|
||||
result = lab046.run_continuous_object_transport(obj)
|
||||
assert result.receive.preamble_candidates == 68
|
||||
assert result.receive.valid_phy_headers == 68
|
||||
assert result.receive.fec_pass_count == 68
|
||||
assert result.receive.crc_pass_count == 68
|
||||
assert result.receive.valid_fragments == 68
|
||||
assert result.receive.object_complete
|
||||
assert result.sha_match
|
||||
assert result.object_byte_errors == 0
|
||||
assert result.stream_plan.soft_symbols.dtype == np.float32
|
||||
assert result.stream_plan.soft_symbols.nbytes < 2_000_000
|
||||
|
||||
|
||||
def test_continuous_background_16_kib_object_is_exact() -> None:
|
||||
obj = lab046.make_deterministic_object()
|
||||
result = lab046.run_continuous_object_transport(
|
||||
obj,
|
||||
leading_background_symbols=37,
|
||||
inter_frame_background={7: 11, 23: 19, 51: 7},
|
||||
trailing_background_symbols=43,
|
||||
seed=464,
|
||||
)
|
||||
assert result.stream_plan.leading_background_symbols == 37
|
||||
assert result.stream_plan.inter_frame_background_symbols == 37
|
||||
assert result.stream_plan.trailing_background_symbols == 43
|
||||
assert result.receive.valid_frame_count == 68
|
||||
assert result.receive.object_complete
|
||||
assert result.sha_match
|
||||
|
||||
|
||||
def test_continuous_noisy_16_kib_object_is_exact_after_fec() -> None:
|
||||
obj = lab046.make_deterministic_object()
|
||||
result = lab046.run_continuous_object_transport(
|
||||
obj,
|
||||
noise_sigma=0.40,
|
||||
seed=46001,
|
||||
)
|
||||
assert result.receive.valid_frame_count == 68
|
||||
assert result.pre_fec_coded_bit_errors > 0
|
||||
assert result.post_fec_protocol_bit_errors == 0
|
||||
assert result.receive.crc_pass_count == 68
|
||||
assert result.receive.crc_fail_count == 0
|
||||
assert result.receive.object_complete
|
||||
assert result.sha_match
|
||||
|
||||
|
||||
def test_continuous_receiver_has_no_tx_oracle_parameters(monkeypatch) -> None:
|
||||
obj, stream_plan = _small_continuous_stream(leading=5)
|
||||
parameters = set(inspect.signature(lab046.scan_continuous_soft_stream).parameters)
|
||||
forbidden = {
|
||||
"frame_plans",
|
||||
"frame_starts",
|
||||
"frame_count",
|
||||
"fragment_indexes",
|
||||
"packets",
|
||||
"coded_bits",
|
||||
"original_object",
|
||||
}
|
||||
assert parameters.isdisjoint(forbidden)
|
||||
|
||||
def forbidden_tx_lookup(*args, **kwargs):
|
||||
raise AssertionError("scanner обратился к TX-side plan")
|
||||
|
||||
monkeypatch.setattr(lab046, "build_object_frames", forbidden_tx_lookup)
|
||||
received = lab046.scan_continuous_soft_stream(stream_plan.soft_symbols)
|
||||
assert received.recovered_object == obj
|
||||
|
||||
Reference in New Issue
Block a user