0 Fork 0

Files

T

mckero f4f6ec7db5 feat(cw): ship the full fp32 DeepCW model instead of the int8 build

用户要求内置完整版模型, 不要量化版。

assets/deepcw/model.onnx: 4,354,478 bytes (int8) -> 15,139,839 bytes (fp32)
sha256 ef120799457bca042d4690944f0faf93268eb4654e7f50f28784ad63bdc1fe02,
与上游 commit 8e264d2 发布的原始文件逐字节一致, 零修改。

实测验证(直接对仓库内的 asset 跑推理, 35 个场景):
- 信噪比: 干净 ~ -6 dB 全部逐字符正确; -9 dB 起显著劣化
- 速度: 12-45 WPM, 8 档中 7 档零错误 (40 WPM 推理仅 167ms)
- 音调: 450-1150 Hz 全窗口 6/6 零错误
- 频率漂移: +20/+60/+150/-300 Hz 全部 4/4 零错误 (卫星多普勒无忧)
- QSB 衰落: 6/12 dB 无损, 20 dB 深衰落 CER 23.5%
- QRM 同频干扰: 4/4 失败(会把干扰台内容一起解出), 全频段模型固有特性,
  实用时依赖电台窄带 CW 滤波器缓解
- fp32 vs int8 准确率打平(5 档中 4 档完全一致); 服务器 x86 上 fp32 推理
  耗时约为 int8 的一半(int8 动态量化的反量化开销在无 int8 加速指令的 CPU
  上反而更慢)。手机 ARM 侧表现待装机确认。

NOTICE.md / DEEPCW.md / README.md 同步更新: 移除 int8 量化派生的记录与复现
步骤, 改为声明未修改照搬上游。

代价: APK 体积约 59MB -> 70MB, 运行内存峰值上升。此前真机闪退的根因是 R8
缺 -keep ai.onnxruntime.** 规则(已修), 与模型大小无关。

2026-08-13 14:48:56 +00:00

3.4 KiB

Raw Blame History

Look4Sat: Satellite tracker

Radio satellite tracker and pass predictor for Android, inspired by Gpredict

Track satellite passes with ease!

Thanks to Celestrak and SatNOGS you have access to over 9000 active satellites.
You can search the entire database by NORAD Catalog Number or the satellite's name.

Orbital positions and passes are calculated relative to your location.
To get reliable data make sure to set the station position via the app Settings.

The application is built using Kotlin, Coroutines, Jetpack Compose and Navigation.
It is now and always will be completely ad-free and open-source.

Main features:

Predicting satellite positions and passes for up to 10 days
Showing the list of currently active and upcoming satellite passes
Showing the active pass progress, polar trajectory and transceivers info
Showing the satellite positional data, footprint and ground track on the map
Custom TLE satellite data import is available via Three Line Element .txt files
Offline first: calculations are made offline. Weekly TLE data update is recommended.

License

The Look4Sat application code is licensed under the GNU General Public License v3.0.

The CW decoder in feature/cw bundles the DeepCW neural decoding model, licensed under the GNU Affero General Public License v3.0 only (AGPL-3.0-only). Because the combined work incorporates an AGPL-3.0 component, the combined work is distributed under the GNU Affero General Public License v3.0 — GPL-3.0 Section 13 permits the combination, and AGPL-3.0 Section 13 applies to the combined work as a whole. Model provenance and attribution are documented in feature/cw/licenses/NOTICE.md; the original GPL-3.0 text is preserved at feature/cw/licenses/Look4Sat-GPL-3.0.txt. The CW model runs locally on-device and does not provide services over a network.

3.4 KiB Raw Blame History