feat(esp32): #83 brownout defense-in-depth (PHY TX backoff + connectivity log)

firmware/esp32-matter/CMakeLists.txt

@@ -2,7 +2,7 @@
 #   . ~/esp/esp-idf-v5.5.4/export.sh && . $ESP_MATTER_PATH/export.sh
 #   idf.py set-target esp32 && idf.py build flash monitor
 cmake_minimum_required(VERSION 3.16)
-set(PROJECT_VER "1.1.7")
+set(PROJECT_VER "1.1.8")
 
 # Unified versioning (issue #77): the Matter softwareVersion INT is DERIVED from PROJECT_VER --
 # MAJOR*10000+MINOR*100+PATCH -> a readable, strictly-monotonic uint32. This keeps the human

firmware/esp32-matter/main/app_main.cpp

@@ -971,6 +971,32 @@ class HisenseOTARequestorDriver : public chip::DeviceLayer::ExtendedOTARequestor
 };
 static HisenseOTARequestorDriver s_hisense_ota_driver;
 
+/* #83 defense-in-depth: Wi-Fi connectivity-transition diagnostics. CHIP's ESP32 ConnectivityManager
+ * already owns Wi-Fi reconnect and auto-re-arms on a FIXED CHIP_DEVICE_CONFIG_WIFI_STATION_RECONNECT_
+ * INTERVAL (5 s) timer (DriveStationState), so the "re-arm after a genuine association loss" half of
+ * #83 is already covered. A runtime exponential backoff would be nicer, but esp-matter's ESP32
+ * platform DECLARES ConnectivityManagerImpl::_SetWiFiStationReconnectInterval WITHOUT defining it
+ * (only _Get links; the member is set once at init), so ConnectivityMgr().SetWiFiStationReconnect-
+ * Interval() does not link on this platform -- and patching the SDK for a marginal win is not worth
+ * it. Note too: the field "never reconnects until power-cycle" symptom is the brownout REBOOT loop
+ * (the device resets at phy_init before CHIP even runs), which no reconnect logic can fix -- only the
+ * bulk cap / custom PCB (#11) does. So this handler stays purely OBSERVATIONAL: it timestamps
+ * Lost/Established transitions on the hisense_ac log so a real Wi-Fi drop can be told apart from a
+ * brownout reboot when triaging a field node. Runs on the CHIP event loop (PostEventOrDie ->
+ * DispatchEvent) with the stack lock already held; reads the event only, touches no CHIP state. */
+static void wifi_connectivity_log_handler(const chip::DeviceLayer::ChipDeviceEvent *event, intptr_t)
+{
+    if (event->Type != chip::DeviceLayer::DeviceEventType::kWiFiConnectivityChange) return;
+
+    const auto result = event->WiFiConnectivityChange.Result;
+    const unsigned up_s = (unsigned) (esp_timer_get_time() / 1000000);
+    if (result == chip::DeviceLayer::kConnectivity_Established) {
+        ESP_LOGW(TAG, "Wi-Fi connectivity: ESTABLISHED (uptime %us)", up_s);
+    } else if (result == chip::DeviceLayer::kConnectivity_Lost) {
+        ESP_LOGW(TAG, "Wi-Fi connectivity: LOST (uptime %us) -- CHIP auto-reconnects on its ~5s timer", up_s);
+    }
+}
+
 static void https_ota_task(void *arg)
 {
     ESP_LOGW(TAG, "HTTPS-OTA: fetching %s", HISENSE_OTA_URL);
@@ -1520,6 +1546,13 @@ extern "C" void app_main()
         esp_matter_ota_requestor_set_config(ota_cfg);
     }
 
+    // #83 defense-in-depth: log Wi-Fi connectivity transitions. Registers an observer for CHIP's
+    // own connectivity-change event so a field node's Lost/Established transitions are timestamped
+    // (tells a real Wi-Fi drop apart from a brownout reboot). Observer only -- CHIP's DriveStation-
+    // State owns the actual 5s auto-reconnect (its interval setter is unimplemented on esp-matter
+    // ESP32, so no runtime backoff). Safe to register any time after esp_matter::start().
+    chip::DeviceLayer::PlatformMgr().AddEventHandler(wifi_connectivity_log_handler, 0);
+
     // HA entity labels (UserLabel key "ha_entitylabel") -> distinguishable same-type entities.
     // start() has returned (Server::Init done, endpoints exist); take the stack lock since we
     // run on a different task than the Matter event loop. Read live by the UserLabel cluster.

firmware/esp32-matter/sdkconfig.defaults

@@ -46,8 +46,8 @@ CONFIG_CUSTOM_DEVICE_INFO_PROVIDER=y
 # so it (and the fallback NUMBER) MUST stay equal to PROJECT_VER or the device reports a stale
 # softwareVersionString. Edit PROJECT_VER in CMakeLists.txt, then update both lines below.
 # esp32-lint.sh enforces this equality (fails the commit/CI if they drift).
-CONFIG_DEVICE_SOFTWARE_VERSION_NUMBER=10107
-CONFIG_DEVICE_SOFTWARE_VERSION_STRING="1.1.7"
+CONFIG_DEVICE_SOFTWARE_VERSION_NUMBER=10108
+CONFIG_DEVICE_SOFTWARE_VERSION_STRING="1.1.8"
 # --- OTA hardening (faster + reliable on marginal Wi-Fi) ---
 # Delta OTA: ship a diff (tens of KB) instead of the full ~1.5MB image over BDX.
 CONFIG_ENABLE_DELTA_OTA=y
@@ -75,6 +75,14 @@ CONFIG_ESP_BROWNOUT_DET=y
 CONFIG_ESP_BROWNOUT_DET_LVL_SEL_0=y
 CONFIG_ESP_BROWNOUT_DET_LVL=0
 
+# Reduce Wi-Fi TX power on the boot that FOLLOWS a brownout reset (#83). IDF-native, no code:
+# esp_phy_reduce_tx_power() forces the lowest PHY TX power before register_chipv7_phy on the boot
+# immediately after esp_reset_reason()==ESP_RST_BROWNOUT, trimming a little current on the retry
+# boot so a marginal rail is likelier to clear re-association. HONEST SCOPE: this is a POST-brownout
+# backoff, NOT a cut to the initial phy_init calibration PEAK that trips the reset (esp-idf#1308) --
+# only the bulk cap / custom PCB (#11) fixes the peak. Belt-and-suspenders, not a hardware substitute.
+CONFIG_ESP_PHY_REDUCE_TX_POWER=y
+
 # Task watchdog PANIC (#12): turn a hung task into a reboot (which re-runs the Wi-Fi/Matter/RS-485
 # bring-up) instead of a warning + a wedged node (Matter + A/C both dark) until a power cycle --
 # exactly the "recover without a reflash" case #12 is about. The RS-485 bus task explicitly