MMB Chapter 101 Bottleneck
Chapter 101 Bottleneck
The delivery and acceptance of the third phase of Sky Dome has entered its final sprint, but a new problem has emerged.
Zuo Cheng was looking at the monthly data of the open platform when he received Tang Xu's call. The voice on the other end of the phone was very serious, and he rushed to the laboratory immediately.
"General Manager Zuo, the efficiency of inter-satellite link scheduling is not meeting the standards." Tang Xu pointed to the simulation curve on the screen. "We ran a full network simulation with 480 satellites, and the spectrum utilization rate of the traditional scheduling algorithm was only 62%, nearly 20 percentage points short of the design target of 80%."
Zuo Cheng frowned as he looked at the data in the simulation report. The spectrum utilization rate was 62%, meaning that the third phase of the Tianqiong satellite network could only operate at 60% of its capacity, with the remaining 40% being wasted. The contract clearly stated that if the spectrum utilization rate fell below 75%, the final payment would be deducted; if it fell below 70%, the project would be postponed.
"Where is the problem?" Zuo Cheng asked.
"The dynamic scheduling of inter-satellite links is too complex," Tang Xu drew a diagram on the whiteboard. "480 satellites are orbiting at high speed, and the link quality between each satellite and its neighbors changes every second. Orbital altitude, satellite spacing, atmospheric interference—these factors combine to create a constantly changing spectrum environment. Traditional algorithms use fixed strategies, allocating resources according to preset priorities, but the environment changes too rapidly, and fixed strategies can't keep up."
"Does it require adaptive scheduling?" Zuo Cheng asked.
"Yes, the algorithm needs to dynamically adjust based on the real-time link status," Tang Xu said. "This is essentially a multivariate optimization problem; there are too many variables for traditional algorithms to handle. We need to use deep learning to solve it, allowing AI to learn to perform optimal scheduling in complex environments."
Zuo Cheng fell silent. AI, it's AI again.
He had anticipated that AI would be the next challenge for 402, but he didn't expect it to come so quickly. If the spectrum utilization rate does not reach 80%, the delivery review will be delayed, and the remaining 120 million in contract payments will be withheld.
"Does anyone here understand deep learning?" Zuo Cheng asked.
Tang Xu shook his head: "The core team of 402 is in communications and IoT, they don't have any AI talent. Ma Hao's machine learning experience is limited to simple prediction models; deep learning is on a completely different level."
Zuo Cheng stood up and walked to the window. On the horizon, a few clouds were moving slowly, as quiet and distant as satellites in space.
He needs AI skills, but 402 doesn't have an AI team. Hiring? Too late. Deep learning engineers can't be recruited in a month or two, and even if they are, they need time to gel. Outsourcing? Impractical; core algorithms can't be handed over to outsiders.
There's only one option left: do it yourself.
Zuo Cheng turned to Tang Xu: "Give me a compilation of all the data and historical simulation results on inter-satellite link scheduling, including the patterns of link quality changes, constraints on spectrum allocation, and the optimization objective function. The more detailed, the better."
Tang Xu was somewhat surprised: "President Zuo, you're going to do it yourself? Deep learning is a very demanding field."
"Let me see the problem size first," Zuo Cheng said. "If the variables are within a controllable range, perhaps a very deep model isn't needed; a shallow network plus an optimization algorithm might be enough to solve it."
Tang Xu nodded: "Okay, I'll get started today. But if we really want to use deep learning, we don't even have a training environment; we'll have to rebuild the GPU servers."
"You organize the data first; I'll figure out the hardware."
After Tang Xu left, Zuo Cheng closed the office door and opened the system panel.
On the technology tree, the AI branch is still grayed out and not yet activated. The activation conditions show that at least three core AI technologies need to be mastered. Currently, Zuo Cheng only masters edge AI inference, and is still missing two.
However, a new notification appears on the system panel: "[Edge AI Inference] The leaf has been associated with the Internet of Things branch and can participate in the fusion."
Zuo Cheng stared at the prompt, his mind racing. If the edge AI inference leaf could be technologically integrated with other leaves, perhaps a more powerful AI leaf could be created, thus meeting the conditions for activating the AI trunk more quickly.
He brought up the technology integration interface, selected "Edge AI Inference" as the main blade, and then browsed other blades that could be integrated with it.
The "Intelligent Adaptive Modulation System" integrated into the communication engineering branch caught his attention. This single component itself is based on an adaptive algorithm using machine learning, which is logically similar to edge AI inference. If the two were combined, it might create an AI scheduling capability tailored to communication scenarios.
Zuo Cheng hesitated for a few seconds, then pressed the fusion button.
A confirmation prompt appeared on the panel: [Technology Integration will consume 5 points. Continue?]
Zuocheng confirmed.
A beam of light shone from the center of the panel, and the images of two leaves began to rotate, overlap, and merge. A few seconds later, the light faded, and a brand new leaf appeared on the panel.
Intelligent Star Network Dispatch System
Type: Fusion blade
Source: Intelligent Adaptive Modulation System + Edge AI Inference
Capabilities: Dynamic scheduling of inter-satellite links based on deep reinforcement learning, supporting real-time optimized allocation of spectrum resources.
Efficiency increase: 1.2 times the technology increase triggered by the fusion solution.
Zuo Cheng took a deep breath. This was what he needed.
The intelligent satellite network scheduling system is based on deep reinforcement learning for dynamic scheduling of inter-satellite links. This is precisely the missing piece of the puzzle for the Sky Dome project.
He checked his points: 205 minus 5, leaving him with 200 points. But the value of this single leaf far exceeded 5 points.
Zuo Cheng closed the system panel, picked up a pen, and began drawing an architecture diagram. He didn't need to write the code himself; he only needed to tell Tang Xu the algorithm framework and key parameters, leaving the engineering implementation to the team. But this time, he needed to spend time understanding the basic principles of deep reinforcement learning; otherwise, he wouldn't be able to provide a correct design solution.
Fortunately, he has a technological advantage. The intelligent star network scheduling system not only provides algorithmic guidance but also enhances the efficiency of the solution by 20%. This means he only needs to design a solution with 67% efficiency, and with the enhancement, he can achieve 80%.
He wants to conquer AI himself. Not for the sake of the tech tree, but for the future of 402.
Zuo Cheng picked up his phone and sent Yu Ying a message: "Kongkong, do you know anything about deep reinforcement learning?"
Yu Ying replied, "I've learned a little, why?"
"Project SkyDome has encountered a bottleneck and may need to use deep reinforcement learning to solve the inter-satellite link scheduling problem. Are there any recommended introductory materials?"
Yu Ying posted a series of links: "These papers are introductory classics. First, read Sutton's Introduction to Reinforcement Learning, then read Mnih's DQN paper. But bro, are you sure you want to try it yourself? This field has a very high barrier to entry."
Zuo Cheng replied, "Let's give it a try."
He put down his phone and looked out at the night sky. Although the city's light pollution obscured most of the stars, he knew that 480 satellites were orbiting the Earth overhead, their links flashing incessantly like an invisible net.
He wants to make this network smarter.