1. Industry background
The past 2022 has been a difficult year for everyone. The difficulty lies in the impact of the epidemic and the slowdown in the growth of the macro environment. No growth brings a painful experience. For the Internet industry, the theme of 21 years is "anti-996". In 22 years, the trend has become "cost reduction and efficiency increase", "business search" and "layoffs". Further refinement of the mobile terminal, after ten years of development, it has entered a "mature period", all walks of life have been transformed, the technical wheels are all there, and the basic services are almost built, it seems We have really reached the bottleneck, and the manpower demand for minor repairs and repairs in the stock era has been cut in half. Maimai's "Beaching the Digital Era Talent Migration Report 2023" report shows that in 2022, the total number of positions recruited by enterprises will decrease by 21.67% year-on-year, and the number of pure Internet positions will decrease by 50.4% year-on-year.
It's time to formulate OKRs for the new year. Everyone is worrying about the planning of technical projects, and they don't know what explorations and breakthroughs to make in technology. The third core conclusion of the "China Software Technology Development Insights and Trend Forecast Research Report 2023" released by InfoQ shows that in 2022, the concept of technical services will change, from advanced technology to business empowerment, and the positioning of the IT department will gradually shift from cost department to business empowerment Departments and technologies have become more marginalized, and personal career development barriers have emerged. At this time, we can't help being confused or even questioning our future. Refinement to the mobile terminal, the frontier and early promotion projects in the "China Technology Maturity Assessment Curve" shown in the "China Software Technology Development Insight and Trend Forecast Research Report 2023" seem to have little to do with the mobile terminal.
This article attempts to explore the possible development directions of the mobile terminal from various aspects, and "squeeze" the possible technical directions to the greatest extent (some just raise questions, not the final answer).
2. Exploration directions and achievements of major manufacturers in the past two years
Before digging, take a look at what the big factories (maybe they have made achievements in a certain field) have been doing these years to see if there is any work that can be directly copied.
2.1 Survey Results in 21 Years
When I wrote OKR in early 21, I did research on several major companies. Let’s take a look at what Ali, Meituan, and JD.com have done and what they plan to do:
Panoramic view of Alibaba's mobile technology
Ali mobile terminal development trend
Meituan Mobile Technology Panorama
Panorama of JD’s mobile technology
JD.com's vision for the future of mobile terminals
It all looks much the same, and probably companies of all sizes are building or are building.
2.2 Output in 22 years
Looking at the output of major manufacturers in 22 years, this mainly comes from the output content of the enterprise technology official account.
2.2.1 Ali
阿里推出的《2022技术人的百宝黑皮书》总结了2022年阿里年度精选终端技术栈的内容：
2.2.2 美团
下面内容摘自美团技术发布的《2022年美团技术年货-合辑》：
2.2.3 百度
百度App技术公众号发布2022精选文章：
2.2.4 分析
从上面三家企业对外输出的文章看，在移动端的动作不外乎几个方向：
- 跨端/低代码
- 性能优化
- 自动化测试
- 开发平台/平台化能力
- AI
3. 移动端主要方向分析
结合上面整理出来的，我们看看移动端“可以”有哪些方向。
3.1 业务开发
业务开发还是主流的市场需求，这块会占大部分的比例，IT部门从成本部门转为赋能部门后，主要的工作量就是支持业务。
3.2 跨端/低代码
在降本增效的背景下，跨端还会持续搞，但是也不是新东西了。H5、React、Flutter、小程序，这些都各有利弊，不同场景用不同技术，像小程序这种更适合平台化的超级APP，规模不够大的话，性价比不高。
3.3 性能优化
同样的，性能优化也是需要持续做的事情，但是也不是新东西了，一些技术手段都比较成熟了，没有太多可挖掘的空间了。
3.4 架构方向
架构管理方向随着规模的收缩，很难出现机会了。
3.5 开发平台建设
在公司内部，类似于蚂蚁的mPaas开发平台在业务快速成长期对提效会有很大的帮助，这个时候随着业务的裂变，推出各种APP，开发平台可以避免很多重复的工作，助力应用快速上线和运营，但是在收缩期再去建设就有点不划算了。
单点的平台能力，比如监控、埋点之类的或者用第三方或者也自建完成了，对缺失的个别能力，可以根据业务需求点滴建设。
3.6 系统应用/Framework/驱动开发
随着AI、Iot、新能源的发展与兴起，释放出一些系统开发的诉求，相对于之前，嵌入式驱动开发的薪资也有所增长，也算是一个方向，但是也要记住，比起手机，电视、汽车毕竟是少数，如果纯转嵌入式的话可能沾Iot的光规模更大些，不过比起芯片，这也是比较成熟的技术，可挖掘方向不大，只是多了个写业务的战场。
3.7 XR
目前比较成熟的是VR，但是VR在端上展示主要基于H5，采集会有单独硬件，有些也支持了手机采集，但是还是那句话，市场需求不大。至于AR、元宇宙更多的是AI的综合应用了，我们也不讨论了。
3.8 音视频
音视频一直是移动端比较大和前沿的一块方向，但是现在也已趋于成熟，下面看看主要的几个方向：
- 点播：播放器的事情，主要涉及多解码期、预加载秒开，剩下的交给系统播放器都可以完成的很好了；
- 录制：系统录制工具，或者基于系统采集、编码、封装封装一套；
- 视频编辑和特效处理：编辑主要是解复用--->解码--->逐帧处理--->编码--->复用的过程，逐帧处理用到视频上主要设计合成、滤镜等；音频主要是变声、声音融合，都是通用的技术，稍微体现差异的就是特效处理中与AI的结合，比如美颜、带眼镜等会用到图像检测，但是都不是门槛，也谈不上前沿探讨；
- 直播：直播也同样有成熟的结局方案：采集推流，开源服务端以及成熟CDN，播放ijk，秒开之类的都是参数优化了；
- 实时音视频：实时音视频开发成本比较高，主要的挑战是弱网对抗，3A处理等，由于不是通用协议，没有CDN，自己搭建机房成本高，而且不见得效果比第三方好，所以也是一件性价比比较低的事情。
- 编解码：目前主流的还是H264，VP8，H264甚至都没有推开，限制编解码算法的主要是推广和兼容性，所以编解码器都是一些组织去搞，一个公司贸然去开发，风险很大。
3.9 AI
人类一直在追求更智能的机器，AI是未来，所以即使现在不够好，并且没有找到太多的落地场景，但是很多公司还在搞，尤其是ChatGPT的能力让大家惊讶，但是它仍然不是真正的“像人类一样”的智能。目前通用的AI主要有一下几个方向：
-
语音方向
- 前端信号处理
- 唤醒
- 语音识别
- 声纹
- TTS
- 作曲：抖音之前分享有过这方面实践和应用
- Feature-based speech coding: such as Google's lyra and SoundStream, Lyra is designed to work at 3kbps, and listening tests show that at this bitrate, Lyra outperforms any other codec and outperforms Opus's 8kbps , thus achieving a bandwidth reduction of more than 60%. But as mentioned above, the bottleneck of the codec mainly lies in the promotion of the standard.
-
image orientation
- detection
- identify
- Image comparison (applied to UI automation testing)
-
natural language processing
- Smart Q&A
- intent recognition
- Document correction
-
Wind control
-
recommend
-
User portrait
-
Metaverse/Digital Human: Digital Human is more like a comprehensive application of AI.
There are also some special users in special special business scenarios, such as the real estate field:
- House type interpretation (based on special features of images)
- Training Course
For AI, what can be explored on the mobile side? To answer this question, we must first understand which scenarios are adapted on the end. Android officially gave a decision-making standard:
The above-mentioned special needs to be applied on the end mainly include:
- wake
- image detection
- voice codec
Can be applied on the terminal:
- ASR
- TTS
- image tag
What is the main workload based on these scenarios? Most of the model training is still placed on the cloud. On the end, it is to load the model, input data, display the output results, and possibly optimize the engine and framework:
4. Summary
On the whole, the development of the entire mobile terminal technology can be said to be "at the end of the mountain". There are not many innovative contents that can be tapped, and most of them are maintained and iterated in the existing system. On the whole, business support is still the main source of demand. In-vehicle machines and Iot have also released some opportunities. Cross-terminal, development platform, performance optimization, and VR have matured, and the implementation of end-to-end intelligence is still in the general direction of voice and image. Combining business has yet to be explored. The current AI still solves the "decision-making" problem. There is still a long way to go from the current generative to the future creative general and "human-level" "intelligence". No one can guarantee that Ray Kurzwe Can the "singularity" of the singularity theory proposed by Err come and when will it come? The evolution of intelligence is not only at the algorithm level, but also affected by computing power. Like the MOSS robot in the "Wandering Earth" series, it is because of quantum The addition of fast computing power.