Horldsence
c7f42a1979
探测脚本是诊断工具，测错比不测更糟——它给出的数字会被当成结论。
**重组状态必须按 (sender_id, frame_id) 存。** 原先只按 sender_id 存，于是丢片
之后下一帧的分片会接着凑上一帧的计数，凑够了交出去一张混着两帧的 JPEG，表现为
"JPEG 不完整"。修完：14 秒 132 帧，0 张坏帧。
换帧时被放弃的半成品单独计为"被新帧取代"，不并入"JPEG 不完整"——它们本来就
不完整，混在一起会把两种情形都算成链路有问题。
**分窗 fps 的分子分母必须同量纲。** 原先用累计帧数除以窗口时长，于是报出 39、
61、79 这种随时间单调上升的假数字，看起来像板子越跑越快。改成窗口增量除以窗
口时长后稳定在 22，也就是传感器上限。
顺带修正两处与代码不符的注释：
partitions.csv / sdkconfig.defaults
两处都写死 app 分区在 0x40000，实际是 scripts/flash.sh 自动挑选并写回，
当前为 0x060000。补上 flash.sh 会重编译并复验的原因（重编译会让镜像位移，
所以偏移无法是常量）。
.env.example
补 CAM_VSYNC_INVERT 和 WIFI_AUTH。WIFI_AUTH 一直能用但没进文档：通常不用
设，扫描会报告 AP 的真实认证方式；只有扫不到时（隐藏 SSID）才需要指定，
指定后只试这一种，不成功直接报错而不是挨个试 10 秒——挨个试等于在猜一个
本该已知的值。
105 lines
4.7 KiB
Plaintext
105 lines
4.7 KiB
Plaintext
# Rust needs more stack than C for the main task and for the threads esp-idf-svc
|
|
# spawns internally.
|
|
CONFIG_ESP_MAIN_TASK_STACK_SIZE=8192
|
|
CONFIG_ESP_SYSTEM_EVENT_TASK_STACK_SIZE=4096
|
|
CONFIG_FREERTOS_IDLE_TASK_STACKSIZE=4096
|
|
CONFIG_PTHREAD_TASK_STACK_SIZE_DEFAULT=4096
|
|
|
|
# Run the CPU at 240 MHz.
|
|
#
|
|
# This is the single biggest win in this file and it is free. ESP-IDF v5.5's
|
|
# Kconfig defaults the ESP32-S3 to *160* MHz (see
|
|
# components/esp_system/port/soc/esp32s3/Kconfig.cpu), so a stock build quietly
|
|
# leaves half the CPU on the table. Camera capture, the JPEG search in
|
|
# esp32-camera, lwIP and the HTTP server are all CPU-bound here, so this moves
|
|
# all of them at once.
|
|
CONFIG_ESP_DEFAULT_CPU_FREQ_MHZ_240=y
|
|
|
|
# ESP32-S3-N16R8: 16 MB flash, 8 MB octal PSRAM.
|
|
#
|
|
# DIO at 40 MHz, not the QIO/80 MHz this part is rated for: the flash on the
|
|
# test module has a defective byte (see README) and quad mode pushes it over
|
|
# the edge into a boot loop. Drop back to QIO/80 MHz on a healthy module.
|
|
#
|
|
# QIO would speed up *code* execution noticeably, since the app runs from flash
|
|
# and the flash cache misses go straight to this bus.
|
|
CONFIG_ESPTOOLPY_FLASHSIZE_16MB=y
|
|
CONFIG_ESPTOOLPY_FLASHMODE_DIO=y
|
|
CONFIG_ESPTOOLPY_FLASHFREQ_40M=y
|
|
|
|
# Code runs from flash, so instruction cache is the cheapest way to hide the
|
|
# slow flash bus. 32 KB instead of the 16 KB default.
|
|
CONFIG_ESP32S3_INSTRUCTION_CACHE_32KB=y
|
|
|
|
# Without this ESP-IDF uses partitions_singleapp.csv, whose factory partition is
|
|
# only 1 MB - smaller than the ~1.15 MB this firmware produces, so the image
|
|
# overflows the partition.
|
|
#
|
|
# The custom table also carries an app offset that keeps the image off the
|
|
# module's defective flash byte. That offset is written by scripts/flash.sh, not
|
|
# fixed here; see partitions.csv for why it cannot be a constant.
|
|
CONFIG_PARTITION_TABLE_CUSTOM=y
|
|
CONFIG_PARTITION_TABLE_CUSTOM_FILENAME="../../../../../../partitions.csv"
|
|
CONFIG_PARTITION_TABLE_OFFSET=0x8000
|
|
|
|
# Octal PSRAM is mandatory for the N16R8 module (R8 == 8 MB OPI PSRAM). The
|
|
# camera frame buffers live in it, so this is what makes VGA possible.
|
|
CONFIG_SPIRAM=y
|
|
CONFIG_SPIRAM_MODE_OCT=y
|
|
CONFIG_SPIRAM_SPEED_80M=y
|
|
CONFIG_SPIRAM_USE_MALLOC=y
|
|
CONFIG_SPIRAM_MALLOC_ALWAYSINTERNAL=16384
|
|
CONFIG_SPIRAM_MEMTEST=y
|
|
|
|
# Keep the WiFi and lwIP buffers in internal RAM.
|
|
#
|
|
# This was previously set to `y` on the reasoning that PSRAM is plentiful, which
|
|
# is exactly backwards: WiFi descriptors and TCP segments are touched on every
|
|
# packet, and internal SRAM is an order of magnitude faster to reach than PSRAM
|
|
# behind the memory bus. Putting them in PSRAM shows up as high latency and poor
|
|
# throughput no matter how much PSRAM is free. The frame buffers, which are large
|
|
# and touched rarely by comparison, stay in PSRAM.
|
|
CONFIG_SPIRAM_TRY_ALLOCATE_WIFI_LWIP=n
|
|
|
|
CONFIG_ESP_WIFI_IRAM_OPT=y
|
|
CONFIG_ESP_WIFI_RX_IRAM_OPT=y
|
|
|
|
# Socket buffers sized for a video frame rather than for a text request.
|
|
#
|
|
# A VGA JPEG is ~20 kB. lwIP's 5760-byte default send buffer splits that into
|
|
# four segments, each one a separate pass through the WiFi driver; 17408 is the
|
|
# lwIP-recommended size for bulk transfers and covers a frame in one or two.
|
|
CONFIG_LWIP_TCP_SND_BUF_DEFAULT=17408
|
|
CONFIG_LWIP_TCP_WND_DEFAULT=17408
|
|
|
|
# Camera component
|
|
CONFIG_CAMERA_TASK_STACK_SIZE=4096
|
|
|
|
# Run the camera task on CPU1.
|
|
#
|
|
# The WiFi task is pinned to CPU0. esp32-camera's capture task spends its time
|
|
# in short, latency-sensitive bursts (DMA completion, the JPEG marker search,
|
|
# the copy into the frame buffer) and gets preempted on a shared core, which
|
|
# lengthens every frame. On a two-core part there is no reason to share.
|
|
CONFIG_CAMERA_CORE0=n
|
|
|
|
# PSRAM DMA is left off deliberately.
|
|
#
|
|
# With it on, esp32-camera writes frame buffers in PSRAM straight from GDMA and
|
|
# then looks for the JPEG end-of-image marker only inside the last DMA node
|
|
# (1 KB here). If the marker lands outside that window the frame is thrown away
|
|
# with "NO-EOI", and on this board that is most of them - after a few frames the
|
|
# driver stops handing out frames at all and never recovers.
|
|
#
|
|
# Off, the driver DMAs into an internal-RAM bounce buffer and copies the frame
|
|
# into PSRAM, which costs one extra memcpy of ~23 KB per frame and searches the
|
|
# whole buffer for the marker. That is the right trade for a working stream.
|
|
CONFIG_CAMERA_PSRAM_DMA=n
|
|
# Largest contiguous DMA chunk the driver may ask for.
|
|
CONFIG_CAMERA_DMA_BUFFER_SIZE_MAX=32768
|
|
|
|
# ESP-IDF 5.3+ links both the legacy driver/i2c.h and the new driver_ng I2C
|
|
# driver, and refuses to run both in one image. esp32-camera picks one of them
|
|
# for the SCCB bus; whichever it picks, the other has to stay out of the image,
|
|
# so the conflict check is skipped here.
|
|
CONFIG_I2C_SKIP_LEGACY_CONFLICT_CHECK=y |