Rebirth: Starting from Lighting Up the Tech Tree

Chapter 21 A Phone Call at Home

On the evening of November 23, Zuo Cheng was running the third round of simulation of the scene adaptation algorithm in the laboratory when his phone rang.

The screen displays "Mom".

He paused for a moment. He had been in this world for almost two months, so busy that he barely had time to breathe, and had very little contact with his family.

Zuo Cheng took off his headphones, walked to the corridor outside the laboratory, and answered the phone.

"Cheng'er, have you eaten yet?" His mother's voice carried a special kind of caution, as if she was afraid of disturbing him.

"I ate, Mom. I just got back from the cafeteria." He lied; in reality, he had only eaten one loaf of bread since noon.

"Your dad wants to say a few words to you."

The voice on the other end of the phone changed; it was muffled, rough, and carried a stiffness that suggested someone who was not good at speaking.

"Second brother, how have you been lately?"

His father called him the second son—he was the second oldest in the family, with an older sister who married and moved to a neighboring city.

"Dad's doing great. He's busy with his graduation thesis and has also collaborated on a project with a company."

"Okay, do a good job."

He paused for two seconds. The father wasn't good at talking to his son; every call was just the same few words over and over. But this time, he hesitated for a moment, then spoke again.

"Don't worry about home, your mother and I are fine. It's just that I've been feeling tightness in my chest and having trouble breathing lately. Your mother insisted I go to the hospital, but I think it's nothing serious, just from working too hard."

Zuo Cheng's fingers tightened. He suddenly remembered that in his previous life, his father suffered a stroke six months after graduation and was rushed to the hospital. Although his life was saved, he needed to take medication for life.

"Did you go see it?"

"I didn't go. Hospitals are expensive, and they don't usually find anything wrong anyway."

"Dad, go take a look." Zuo Cheng's voice was low, but his tone left no room for negotiation. "Don't skimp on this money, I'll transfer it to you."

"No need, no need, you're just a student, where would you get the money—"

"I have enough," Zuo Cheng interrupted him. "I received a payment for a project I did with the company; it's enough."

There was silence on the other end of the phone. After a while, the mother's voice came through again, clearly having snatched the phone back.

"Cheng'er, your dad's just stubborn, don't take it to heart. We're not short of money, just focus on your studies. By the way, you'll be graduating soon, right? Have you decided where you want to work? Your Uncle Zhang's daughter graduated this year and works at a bank back home, why don't you—"

"mom."

"Don't rush to refuse, Mom was just saying. You're not getting any younger, it's time to think about your personal life. No matter how good your studies or job are, it won't work if you don't have someone who cares about you—"

"Mom, I understand. This isn't urgent. Can we talk about it when I go back for winter break?"

"Okay, remember, you must come back during winter break. Your sister also said that the whole family will have a reunion dinner together during the Spring Festival, and she'll be the first to object if you don't come back."

"I'll go back, I'll definitely go back."

After hanging up the phone, Zuo Cheng stood in the corridor for a long time.

The corridor lights were voice-activated; he didn't speak or move, and the lights went out. In the darkness, only the lights of the distant dormitory buildings outside the window remained, arranged in grids like a chessboard.

The suffering my father endured in his past life must not be repeated in this one!

Zuo Cheng took out his phone, opened his bank app, and transferred 50,000 yuan home. The note read "Project Bonus".

Fifty thousand yuan is not a small amount for him now—the money in the company's account cannot be touched, the working capital of 402 needs to be kept for operation, and this money is all the savings he has accumulated in the past few months, plus two points he redeemed.

But he didn't hesitate for a second.

After transferring the money, he sent his mother a WeChat message: "Mom, the money has arrived. Take Dad to the hospital for a full physical examination, the most comprehensive one. Don't let him refuse; just say it's my request."

His mother replied with a long string of voice messages, the gist of which was, "Where did you get so much money?" "Did you borrow it?" "Don't spend money recklessly." Zuo Cheng replied to each message one by one, patiently explaining everything.

It was almost ten o'clock when he finished. He returned to the lab, sat down, and stared blankly at the simulation data on the screen for a while.

Chen Hao glanced at him, asked nothing, and silently got up to make him a cup of tea.

Zuo Cheng took a sip of tea, pulling his thoughts back. He could think about family matters later; right now, there was a major problem stuck in the third stage.

The framework for the scene adaptation algorithm has been established, and the specific optimizations for the two scenarios—dense urban areas and highways—have been successfully tested. However, the third scenario—underground space—has encountered problems.

The signal environment underground is completely different from that on the surface. After being attenuated by layers of concrete and steel bars, the signal strength is extremely low and the noise ratio is extremely high when it reaches the terminal. What's more troublesome is that the multipath effect in underground space is extremely complex—the signal is repeatedly reflected and refracted in the confined space, and when it reaches the receiver, it is superimposed into a mess. Conventional channel estimation methods simply cannot distinguish which is the useful signal and which is reflected clutter.

Zuocheng's adaptive tracking algorithm performs drastically in this environment—it's not that it's unusable, but its accuracy is just at the level of Lanwan Communication's existing solutions, unable to surpass them.

A 30% performance improvement is the key performance indicator for stage three. Ground-based scenarios easily exceed the target, while underground scenarios just break even, averaging around 20%—not enough.

"The problem lies in feature extraction." Chen Hao's classifier only achieved 78% accuracy in underground scenes, far lower than the over 95% accuracy in above-ground scenes. "The signal features in underground spaces are too blurry, and the classifier often misidentifies them as indoor coverage scenes, leading to the loading of incorrect parameter configurations."

"Essentially, it's still a matter of insufficient data." Zuo Cheng rubbed his temples. "We have too little actual measurement data of underground spaces; we only have a little bit of marginal data collected occasionally during the 72 hours of base station data collection. The classifier's training set is insufficient, so naturally, it can't make accurate judgments."

There isn't enough data.

This is a problem that cannot be solved by technical means—he cannot conjure up actual channel measurement data for underground spaces out of thin air. Lanwan Communications' experimental base station coverage area does not include subways or underground shopping malls; to obtain data, they would need to find base stations with underground scene coverage.

"Can Han Zhe coordinate data from other base stations?" Fang Ze asked.

"We can give it a try, but the process will take at least a week." Zuo Cheng glanced at the countdown for step three—there were still twenty-six days left. A week's wait wasn't fatal, but it would compress the subsequent debugging time.

He picked up his phone and sent Han Zhe a brief message explaining his data needs.

Then I leaned back in my chair and stared at the ceiling for a minute.

We can't just wait. We need to find a backup plan.

"Chen Hao."

"Um?"

"Let's try a different approach—since there's not enough measured data, let's supplement it with simulated data. Can you generate a batch of high-quality simulated data based on existing underground space channel models? It doesn't need to be exactly the same as real data, but it needs to cover a sufficient number of scene variations to support the training set of the classifier."

Chen Hao adjusted his glasses, thought for a few seconds, and said, "Data augmentation? It's possible, but there's a distributional bias between simulated and real data. Directly mixing them for training might actually lower the classifier's accuracy."

"Then let's not just mix them directly." Zuo Cheng drew a simple diagram on the paper. "We'll use a two-stage training method—first, pre-train with simulated data to let the classifier learn the basic characteristic patterns of underground space; then fine-tune with a small amount of real data to correct distribution biases. This is the idea behind transfer learning."

Chen Hao stared at the simple diagram for five seconds, then slowly nodded.

"I can do this. Give me three days."

Zuo Cheng patted him on the shoulder without saying much.

Three days.

If Chen Hao's data enhancement plan works, the weakness in underground space can be addressed. If Han Zhe can also coordinate access to accurate data at the same time, the chances of success will be even greater.

He glanced out the window. The night was already quite cold at the end of November, and the laboratory's heating made him feel sleepy, but he wasn't sleepy at all.

Half of my mind was racing with the technical details of the scene adaptation algorithm, while the other half was thinking about what my father had said on the phone: "I feel a tightness in my chest."

I'll definitely go back home during winter break. I'll take him for a medical checkup to find out exactly what's wrong.

Technology can be developed gradually, and a company can grow slowly, but some things cannot wait.