On March 1, at the 2022 Mobile World Congress, Guo Ping, Huawei's rotating chairman, said that Huawei is working hard to achieve three reconstructions, including basic theoretical reconstruction, architecture reconstruction, and software reconstruction, which will support the long-term sustainable development of the ICT industry.
Guo Ping said that digitalization and carbon neutrality are two important issues in today's world and have a far-reaching impact on the future of ICT. The global digital economy is developing rapidly, and it is predicted that the digital economy will account for more than 50% of GDP this year, and the demand for digitalization will exceed expectations. On the supply side, Shannon's theorem and von Neumann's architecture have encountered major bottlenecks. Therefore, we need to explore new theories and new architectures to support digital sustainability.
The basic theory is reconstructed, approaching Shannon to break through Shannon
Guo Ping said: "The density of the connection multiplied by the precision of the calculation is the strength of the digital economy. But now we don't just need strength, we need long-term vitality. Therefore, we need to consider one more dimension, that is, the strength of carbon emission reduction. To achieve this goal, Huawei will significantly increase its strategic investment in root technology, and we will work with partners to reconstruct the technical base. ”
Guo Ping said: "We know that the channel capacity is close to the ceiling. Huawei continues to explore theories and technologies such as a new generation of MIMO and wireless AI to further approach the Shannon limit, and at the same time research new theories such as semantic communication, trying to surpass the Shannon limit and open up a broader development space for communication.
Shannon's theorem believes that in a channel, that is, in a certain range of spectrum resources, the amount of data that can be transmitted is limited, from the first generation of mobile communication to the fifth generation of mobile communication, the way of channel coding is changed and changed, the purpose is to improve the utilization efficiency of the spectrum, close to the Shannon limit, and now it is very close to the limit value of the Shannon curve, and the industry continues to develop facing the ceiling of the theoretical basis.
Therefore, Huawei has developed a new generation of MIMO technology and wireless AI technology in terms of software algorithms to further improve the transceiver efficiency of wireless devices and approach the shannon limit. Wireless AI technology reconstructs the physical layer design, channel decoding and signal detection, and intelligent scheduling of wireless networks through artificial intelligence, improving the transmission efficiency of wireless signals and coping with future waves of wireless data.
To break through Shannon's theorem, it is necessary to develop new theories, such as semantic communication.
Semantic communication is popular, just like the secret agent password in the TV series, sending very few bytes and transmitting complex content. Relying on the local semantic knowledge base, the two sides of the communication only transmit a small amount of information, and each of them is coded and decoded separately to transmit complex and huge information. Facing the future, after entering the metaverse era, there will be a large number of videos, a large number of AR VR, especially ar VR in the game, may be a rendering, a light knife screen, originally need to transmit a lot of information, but semantic communication can be realized, at one end of the transmission to send a simple semantic information said that there needs to be a light, the receiving end or the decoding end, it is based on the semantic knowledge base, you can directly render and present the light.
Original, Semantic Encoding Reconstruction, Existing Encoding Reconstruction (H.264 Code)
It can be seen that semantic communication can greatly reduce the amount of data transmitted by the traditional communication channel, but based on the semantic knowledge base, the expressiveness after decoding is still rich. In the future, Huawei may break through and surpass the framework of Shannon's information theory from the direction of semantic communication and develop into a broader space.
Architectural refactoring, breaking through the von Neumann architecture
Guo Ping said that wireless communications are still facing major technical challenges such as high frequencies, ultra-large bandwidth, and ultra-high speed, and Huawei is actively exploring new technologies to reconstruct the architecture, such as introducing optoelectronic fusion technology, solving key problems, and breaking through the process bottlenecks faced by future chips.
Huawei is a world-class technology level in the field of wireless and optical, optical technology can naturally be applied to high frequencies, and some technical problems faced in the evolution of wireless networks to large bandwidth, high frequency bands, and ultra-high speed directions can be achieved through the integration of wireless and light.
The spectrum of wireless networks is getting higher and higher, from the 1GHz frequency below the 2G period to the 2GHz frequency below 3G, the 3GHz frequency below 4G, and the 6GHz frequency below 5G. The benefits of high frequencies are large bandwidth and high rates, which place increasing demands on the processing power of wireless devices such as antennas and base stations, and optical and electrical fusion processing technology can meet such requirements and exceed the limitations of existing architectures.
In the future wireless AAU RF part and BBU baseband part, a large number of chips will have to do complex signal processing, and ordinary electrical chips will be difficult to complete processing. The frequency band of the sixth generation of mobile communications, experts generally believe that the high frequency band of millimeter wave will be used, and even reach terahertz above the frequency of tens of GHz, and traditional radio frequency technology is difficult to support, because there is no way to achieve digital-analog conversion. The spectrum of light itself is very high, so the natural processing of high-frequency wireless signals will have great advantages, and analog video signals can be directly illuminated.
For example, with the wireless rate getting higher and higher, the traditional CPRI interface is difficult to meet the requirements, ROF (Radio Over Fiber) uses analog signals directly transmitted on the light, which can simplify the base station architecture and reduce energy consumption, which is a potential technology supporting the evolution of CPRI.
Guo Ping said that the current contradiction of computing architecture is the vigorous development of AI and big data applications, while the traditional computing architecture is still "CPU-centric". In order to solve this contradiction, Huawei is designing a "peer-to-peer" architecture so that GPUs, NPUs, etc. can better support the development of global AI services.
"CPU-centric" is the Von Neumann architecture that has long dominated the design of CPU microarchitectures, and the most important feature of this architecture is the separation of computing and storage functions. At present, a large number of new applications are emerging, AI, big data is the center of computing, and the traditional CPU that is good at processing integer and floating-point operations cannot handle such applications well, while GPUs, NPUs and new hardware cannot directly access memory and storage. Therefore, Huawei's proposed reconstruction of the computing architecture is to upgrade from the original CPU-centric architecture to a peer-to-peer architecture, so that the main chips such as GPUs and NPUs that are better at running AI software can be fully effective and support a large number of AI software operations.
In the traditional architecture, the computer has a CPU cpu, all tasks are issued by the CPU to the various processors to perform, and Huawei is designing an AI-centric "peer-to-peer" computing architecture, which can break the CPU bottleneck, weaken the cpu's central position, and connect the memory and hard disk through the new architecture to give full play to the performance potential of each processor. This architectural innovation can solve the huge pressure brought by a large number of AI applications and big data applications on computing architecture.
Software refactoring, to achieve AI-centric full-stack software refactoring
"Facing the future, with the outbreak of AI, the demand for computing power has increased sharply, but the progress of hardware processes has slowed down." Guo Ping said, "To this end, we have proposed a 'software performance multiplication plan', such as: the number of wireless cells and scheduling users and other key indicators have been doubled through software optimization; we will more effectively play the computing power potential of diversified hardware through Hongmeng and Euler; through the Mindspore framework, help scientists and engineers improve development efficiency." ”
At present, the demand for AI is growing explosively, and hybrid computing with AI as the center and combining big data and HPC is becoming the mainstream of applications. Huawei's software refactoring is to realize AI-centric full-stack software refactoring.
At the operating system level, Huawei can directly connect a variety of computing hardware in Kunpeng and Ascend through the two operating systems of Euler and Hongmeng to give full play to their respective computing power. Through these two operating systems, tasks can be distributed very quickly. Through software reconstruction from top to bottom, Huawei can support a large number of apps in the future, especially the explosion of AI and big data.
At the software tool layer, large-scale heterogeneous parallel computing, such as AI and big data parallelism, AI and HPC parallelism, is solved through computing frameworks, allowing developers to ignore the complex relationships between multiple frameworks used by AI, HPC, etc., and it becomes easier to develop AI. For the development of AI, Huawei's Mindspore aims to provide data scientists and algorithm engineers with a design-friendly, efficient development experience, and on the basis of overall support for mainstream Pytorch and Tensorflow in the industry, focus on improving Mindspore and making development more efficient.
Guo Ping said that we know that only products with full coordination of software and hardware can really bring a good user experience. We are also practicing this philosophy in ICT product development. For example, the refinement of the base station AHR Turbo algorithm has enabled MetaAAU to achieve performance and energy saving; after the breakthrough of the hologram optical algorithm, OXC has achieved a full-light "one-jump direct". Huawei's continuous investment in root technology will gradually be reflected in product competitiveness.