Even Demo App

The application uses a hybrid architecture where the Flutter (Dart) layer communicates with platform-specific native BLE implementations (Android/iOS) via MethodChannel and EventChannel.

• Downlink (Dart → Native): Uses MethodChannel("method.bluetooth") to invoke methods like send, startScan, or connectToGlasses.
• Uplink (Native → Dart): Uses EventChannel("eventBleReceive") (Android) or blueInfoSink (iOS) to stream binary notifications and business data.
• Status Callbacks: Native code uses methodChannel.invokeMethod to trigger state changes in Dart (e.g., setMethodCallHandler).

// Dart side entry point
// lib/ble_manager.dart

// Example of calling a native method
await methodChannel.invokeMethod("send", {"data": data, "lr": lr});

// Example of listening to incoming data
eventChannel.receiveBroadcastStream().listen((data) => _handleReceivedData(data));

The Even AI feature follows a specific sequence of Bluetooth interactions between the app and the glasses:

1. Activation: After dual Bluetooth connection, the user long-presses the left-side TouchBar. The glasses send command [0xF5, 0x17] to the app.
2. Microphone Activation: The app must respond by sending command [0x0E, 0x01] to activate the right-side microphone.
3. Audio Capture: The glasses stream real-time audio in LC3 format. The user holds the TouchBar while speaking (max 30 seconds).
4. Processing: The app converts the LC3 audio to text and sends it to a Large Language Model (LLM).
5. Result Transmission: The app sends the LLM response to the glasses via the Bluetooth protocol.
   • Automatic Mode (Default): Results are transmitted page by page.
   • Manual Mode: Triggered by a single tap on the TouchBar. Use the left TouchBar for page-up and the right TouchBar for page-down.
6. Exit: A double-tap on the TouchBar exits the Even AI function.

To manage the G1 connection from the Flutter layer, follow this lifecycle using the BleManager class:

1. Scan: Call BleManager.startScan() (invokes startScan via MethodChannel).
2. Handle Pairing: Listen for foundPairedGlasses via the method channel to receive the leftDeviceName, rightDeviceName, and channelNumber.
3. Connect: Call connectToGlasses(deviceName). Note that deviceName should be passed as Pair_{channelNumber}. The native layer will strip the Pair_ prefix.
4. Connection Established: Listen for glassesConnected. Once received, set isConnected to true and start the heartbeat via startSendBeatHeart().
5. Data Reception: Listen to the eventBleReceive EventChannel to handle incoming binary data and commands (e.g., touchpad or AI events).

// Example conceptual flow in Dart
await bleManager.startScan();

// When paired glasses are found
// native calls 'foundPairedGlasses' with {leftDeviceName, rightDeviceName, channelNumber}

await bleManager.connectToGlasses('Pair_45');

// Listen for data
eventChannel.stream.listen((event) {
  _handleReceivedData(event);
});

Image transmission supports 1-bit, 576*136 pixel BMP images. The process requires three steps:

1. Packetization: Divide the BMP data into 194-byte packets. Each packet must be prefixed with the 0x15 command and a syncID (the packet index).
   • First Packet: Must include the glasses' end storage address [0x00, 0x1c, 0x00, 0x00]. The format is [0x15, index & 0xff, 0x00, 0x1c, 0x00, 0x00, ...data].
   • Subsequent Packets: Do not include the address; use [0x15, index & 0xff, ...data].
   • Packets can be sent to the left and right BLE sides independently.
2. End Command: After the last packet is sent, send the packet end command [0x20, 0x0d, 0x0e] to the dual BLE.
3. CRC Check: Once the end command is acknowledged, send a CRC check command using command 0x16. The CRC must be calculated using Crc32Xz big endian, incorporating both the BMP picture storage address and the picture data.

// First packet structure example
[0x15, index & 0xff, 0x00, 0x1c, 0x00, 0x00, ...data]

// Packet end command
[0x20, 0x0d, 0x0e]

The Even AI response flow follows a specific sequence to ensure the glasses can display text correctly:

1. Speech Capture: toStartEvenAIByOS() starts the microphone and listens for speech via an EventChannel named eventSpeechRecognize.
2. Processing: When recordOverByOS() is called, the captured combinedText is sent to ApiDeepSeekService.
3. Text Chunking: The response text is broken into a list of strings using EvenAIDataMethod.measureStringList, which calculates line breaks based on a maximum width (default 488) and font size (21).
4. Paging Protocol:
   • The service first sends a packet with status: 0x30 to prepare the glasses.
   • It then sends a packet with status: 0x40 to trigger the actual display.
   • For long responses, a Timer periodically sends chunks of text every 5 seconds using status: 0x30 until the end of the list is reached, where it finishes with status: 0x40.
5. Manual Override: If the user interacts with the TouchBar, the timer is cancelled, and the service uses status: 0x50 to send manually requested pages.