Rebirth: Starting from Lighting Up the Tech Tree

Chapter 37 Milestone

May 20th, five days before the first milestone acceptance.

Chen Hao's signal reception and demodulation module has been running on the embedded platform, but Fang Ze discovered a problem when doing system-level integration testing—when the demodulation module and Zuo Cheng's channel estimation module run at the same time, the CPU utilization rate will soar to 92%, leaving almost zero computing power for the beam management interface and system processes.

"It's not that the code is badly written, it's that the two modules are competing for resources." Fang Ze sent a long message in the work group at 2 a.m., along with a real-time CPU utilization graph. "The FFT operation of the demodulation module and the matrix operation of the channel estimation module are both computationally intensive tasks, and their peak loads happen to overlap within the same time window. They run fine individually, but crash when they're integrated."

Zuo Cheng saw the message at 2:10 a.m. He had just returned to his dormitory from the office and was getting ready to wash up and go to bed.

After reading Fang Ze's analysis, his drowsiness vanished instantly.

This problem doesn't exist on PCs—PC CPUs have sufficient headroom to handle the peak loads of two modules simultaneously. However, embedded platforms have limited ARM processor performance; computing power is a hard constraint that cannot be solved by upgrading hardware.

"Let's meet at eight tomorrow morning," Zuo Cheng replied in the group chat. "Chen Hao and Fang Ze, you two should each think of a plan tonight, and we'll discuss it together tomorrow."

Chen Hao replied instantly with an "OK". Fang Ze replied, "I'm already thinking about it."

The next morning, the three of them stood in front of the office whiteboard for two hours.

Chen Hao's solution was to reduce the FFT precision of the demodulation module—using a shorter FFT length to reduce computation. However, this would sacrifice demodulation performance, and Blue Bay Communications' acceptance criteria had specific requirements for bit error rate; reducing precision might not meet the requirements.

Fang Ze's solution involves task scheduling—staggering the peak loads of the two modules and using a time-division multiplexing scheduler to alternately allocate CPU resources. The idea is feasible, but the overhead and latency introduced by the scheduler itself require precise control, making implementation quite challenging.

Zuo Cheng stood in front of the whiteboard, staring at the dispatch diagram drawn by Fang Ze, and the "feel" of the fusion blade in his mind began to heat up again.

"Fang Ze's approach is correct, but a universal scheduler isn't needed." He picked up a pen and drew a timeline on the whiteboard. "The demodulation module's FFT operation has a fixed periodicity—it's executed once per symbol period, and the execution time is predictable. The matrix operation of the channel estimation module is also periodic, but the period is longer. There's a natural 'gap' between the two periods—a free window of approximately 0.3 milliseconds after the FFT ends and before the next symbol period begins."

He marked that gap on the timeline.

"The matrix operations for channel estimation are broken down into small blocks, with the computation time of each block controlled within 0.3 milliseconds, and executed within the intervals of the FFT. No scheduler is needed; it relies on the timing coordination between the two modules—the demodulation module sends a semaphore after completing one FFT, and the channel estimation module, upon receiving the semaphore, begins executing a small block, releasing the CPU upon completion."

Fang Ze stared at the whiteboard for ten seconds.

"Pipeline insertion," he explained, revealing the essence of the design: "Interlocking the computational pipelines of two modules, allowing them to rotate alternately like gears."

"Yes." Zuo Cheng nodded. "This way, the peak CPU utilization won't exceed 75%, and there's no need to introduce additional scheduler overhead and latency."

"But there's a trick to splitting matrix operations into blocks." Chen Hao frowned and thought for a moment. "Splitting it too finely will increase the number of context switches, while splitting it too coarsely will exceed the 0.3 millisecond window. How do you determine the optimal block size?"

"No need to determine it manually." Zuo Cheng wrote down a formula. "The dimension of the channel estimation matrix is ​​known, the symbol period of the FFT is also known, and the optimal block size can be calculated directly—it's a function of the matrix dimension and the symbol period. I derived it last night; the formula is here."

He finished writing the formula and took a step back.

Chen Hao and Fang Ze exchanged a glance.

"Brother Cheng, what time did you go to sleep last night?" Chen Hao asked.

"3:30."

"No wonder." Fang Ze adjusted his glasses. "Alright, I'll implement this plan. Give me two days."

Fang Ze took a day and a half.

On the afternoon of May 22nd, the joint debugging test was run again. Peak CPU utilization was 71%. There was no loss in demodulation performance or channel estimation accuracy. The computational pipelines of the two modules rotated precisely and alternately like two sets of meshing gears, without any conflicts.

Fang Ze looked at the stable CPU curve on the monitoring screen and smiled, a rare occurrence for him.

"pretty."

On May 24th, all deliverables for the first milestone were ready—a working software version, test reports, and technical documents, packaged into an encrypted compressed file and uploaded to Blue Bay Communications' project management platform.

Han Lu simultaneously submitted the milestone acceptance application and the corresponding business documents.

"The acceptance process usually takes three to five business days," she said in the group chat. "We expect to have results by the end of May."

Zuo Cheng replied with "Received" and then closed the work group.

My phone vibrated. It was a message from Zhang Lei, not from a work group, but from a private group chat for the four roommates in room 402.

"Have you heard? Ma Hao is gone."

Zuo Cheng opened the message.

"What do you mean?" Chen Hao asked.

"I ran into another of Professor Lin's graduate students at the lab today. He said that Ma Hao completed his graduation procedures last month and graduated early."

"Where did you go?"

"I don't know. But I heard he got an offer from Shenzhen, something related to chip manufacturing. I haven't found out which company it is."

Zuo Cheng looked at the message but remained silent.

Ma Hao left. He graduated early and left Huaxia City.

In his previous life, Ma Hao stayed in academia, eventually becoming a lecturer at a second-tier university, where he languished in obscurity for over a decade. In this life, after Zuo Cheng exposed his true colors, his academic path was essentially blocked—Lin Zhiyuan was unlikely to give him a good recommendation letter, and many people in the academy knew about it. Choosing to work in the corporate world was the best path for him.

Shenzhen makes chips. Zuo Cheng mentally reviewed the list of chip companies in Shenzhen, but couldn't think of anything particularly relevant.

Never mind, it's not important. Ma Hao's path is his own business.

He closed the private group and reopened the code editor for the channel estimation module.

The first milestone was met, but the even more demanding second milestone was yet to come—the complete version of the channel estimation and beam management module needed to be delivered by the seventy-fifth day. There were still forty days left, but Zuo Cheng knew the real challenge wasn't writing the code, but achieving millisecond-level response on an embedded platform while maintaining uncompromised prediction accuracy.

This is precisely where the "feel" that the fusion blades bring him can play its most significant role.

At nine o'clock in the evening, he came out of his office and walked around the campus for a while.

By the end of May, Huaxia City already had a summery feel; the evening breeze was warm, and tiny insects flew under the streetlights. The sycamore-lined avenue was sparsely populated, with only the occasional couple walking hand-in-hand.

Zuo Cheng's phone rang.

Automatic notification emails from the Blue Bay Communications project management platform.

He clicked on it and saw—

"Project Sky Phase 2, Milestone 1, Acceptance Result: Passed. Acceptance Score: 93/100. Acceptance Comments: The signal reception and demodulation module is fully functional, its performance meets contract requirements, the code quality is excellent, and the documentation is standardized. It is recommended that robustness in multi-mode switching scenarios be considered in future development."

passed.

Faster than expected—Han Lu said three to five working days, but it only took two. Ninety-three points, not amazing, but very solid. The "robustness of multi-mode switching" mentioned in the acceptance comments is indeed a key area they will focus on next.

Zuo Cheng sent a screenshot of the email to the 402 work group.

"First milestone achieved."

The group chat exploded again.

Zhang Lei's first message was: "Brother Cheng said he'd treat us to dinner after we passed this milestone!"

Liu Wei added, "I want to eat a whole roasted lamb."

Fang Ze sent a message: "Let's plan for the second milestone first; there's no rush to eat."

Chen Hao: "I agree with Fang Ze. But I also agree to have dinner."

Zuo Cheng smiled and shook his head, then typed in the group chat: "Saturday night, it's on me. You guys decide what to eat."

Then he closed the work group, stood on the sycamore-lined road, and looked up at the sky.

The night sky in May was not clear enough; the stars were blurred by the city lights, and only a few of the brightest ones could be seen.

But he knew that at those unseen heights, the first experimental satellite of the Celestial Constellation was flying silently along its orbit.

He's writing code for that satellite.

This thought sent a surge of warmth through his chest.

Zuo Cheng withdrew his gaze, took out his phone, and sent a message to Yu Ying.

"The milestone has been reached."

Yu Ying's reply came quickly.

"Congratulations. So, is it almost time to settle that matter you owe me?"

Zuo Cheng looked at the words and couldn't suppress the smile on his face.

"Soon. June 4th."

"good."

Just one word.

But Zuo Cheng felt that this was the most important message he had received today.