Chapter 605: It is not recommended to use too much?
When the first sound of "wrong calculation" came out, some people felt a "thump" in their hearts. Oh no!
But as more and more people said "correct", everyone looked at the few comrades who said "wrong" with amusement in their eyes.
So what, comrade, did you make a mistake in your calculation?
Several comrades looked stern and very embarrassed.
But at this moment, everyone's attention was no longer on whether their calculations were correct, but on the circuit board full of cannulas.
Baby!
Although the significant digits of a 16-bit floating point number are only 3 to 4 digits, and the numerical range is not large, for many practical engineering applications, this accuracy and numerical range are actually sufficient.
Let's take the most typical and widely used example, pi.
Due to the existence of the famous song "One temple on the top of the mountain, one pot of wine", I believe many people can easily memorize Pi to 5 decimal places. Those who are leisurely can easily memorize Pi to 20 digits or more. Of course, "He is Memorizing pi backwards!” This kind of joke is impossible.
However, in actual use, no matter how long it is memorized, a lot of people actually use 3.14. The reason is very simple. The relevant results calculated using 3.14 and 3.14159 have very little difference in engineering applications, and are generally sufficient.
If you do the math, you will know that even if Pi is used to 10,000 decimal places, the difference from 3.14 is only about five ten thousandths.
This is why Gao Zhendong first built a 16-bit floating point hardware multiplier - it is sufficient and cheap.
"General Engineer Gao, what's the name of this board?" A comrade working on radar asked excitedly.
Gao Zhendong smiled and said: "Single-cycle 16-bit floating point hardware multiplier."
This name is very direct, and the comrades understood it immediately. Some of them, who know a lot about computers, exclaimed: "Can this machine complete multiplication operations in one cycle?"
Gao Zhendong nodded: "Yes, its instruction cycle is the same as that of DJS-59. Four clock cycles form one machine cycle. Theoretically, even if the computer code of the DJS-59 series is optimized, it can complete an 8-bit integer The time required for multiplication is more than 25 times that of this hardware multiplier to complete a 16-bit floating point multiplication. In fact, since the main frequency of this device is 1MHz, compared with the 0.6M DJS-59 and the 0.8M DJS-60D , its speed is even faster than what I just said.”
Better integrated circuit technology made the main frequency of the hardware multiplier faster than the first two general-purpose computers from the beginning, and had greater potential.
If converted, the theoretical limit computing power of this hardware multiplier is about 0.25 MIPS, which is 250,000 times per second.
1/25, even if the difference in main frequency is not taken into account, this ratio makes all radar comrades excited.
"This means that we can calculate signals faster!"
"Great, I estimated that with this, the speed of our moving target indication algorithm will be usable."
Although their moving target indication uses the delay line cancellation filter suggested by Gao Zhendong, which can be implemented with analog devices, this does not mean that they do not need mathematical calculations when processing targets.
"This is very useful for our digital communications. Signal processing, encryption and decryption, and the real-time voice signal processing and real-time encryption problems that have always troubled us have finally seen the light of day."
This is a comrade who was specially called by the committee chief engineer to do communications.
However, the matter was not over yet. Gao Zhendong said something that made them even happier.
"Actually, the main frequency of this circuit has not reached the limit. The main frequency I am using now is relatively conservative, and with the existing integrated circuit technology, the main frequency that the device can achieve should be increased several times. ”
PMOS can reach a main frequency of several MHz, which is also the limit under current conditions. Although PMOS is limited by its own principle flaws, it is 1,000 times worse than the speed of several GHz that Gao Zhendong could achieve in his previous life. But in this era, a lot of things are done quickly.
The happiest thing is the comrades who are engaged in artillery position reconnaissance radar. In the algorithm Gao Zhendong gave them, the real-time point position of artillery shells is calculated, and the extended Kalman filter is performed based on the calculated point data set, as well as the Taylor stage that needs to be used. Number expansion, weighted average, etc., all use a lot of multiplication operations. The originally used DJS-60D was a bit limited in calculations due to bit width and architecture issues. Simply put, it was a bit slow.
It is difficult to process the shell point data sent back by the radar in real time. They are now using the method of first storing the radar signal data, then calculating the point data based on it, and finally filtering and positioning based on the point data set. way.
In fact, the time required for the entire process was measured in minutes. By the time the counterattack artillery group had bound the data and sent it back, it would already be several minutes later. It's not that it can't be used, but it's obviously not able to achieve the effect they want.
After all, according to what they know, the most advanced enemies have already used self-propelled artillery. This kind of artillery is capable of running away within tens of seconds after firing several rapid rounds.
Even if it is a towed artillery, in the hands of our skilled soldiers, it is not impossible to retreat on the trailer within a few minutes after firing.
Therefore, compressing the decoding time is their biggest wish. It is unrealistic to pile up DJS-60D, it is too expensive, and the programming is complicated. Now, this wish may come true.
"General Engineer Gao, do you have many of these? Give us one to try."
Gao Zhendong smiled and said: "There is no excess now, but don't worry, this thing is completely domestically produced. You can have as much as you want. But I don't recommend that you use too much."
Is it not recommended to use too much? What do you mean?
All the comrades who work on radar are gearing up and mentally thinking about how to use this thing.
Convolution, series expansion, Fourier transform, Laplace transform, z transform, curve fitting...
There are so many places where it can be used. Just this small circuit board can drastically improve the back-end performance of the radar by at least an order of magnitude!
And then you told us that it’s not recommended to use too much of this thing? Appetizing, right?
Even the cheerful-looking leader of the National Defense Work Committee was stunned by Gao Zhendong's words.
"Zhendong, are there difficulties in producing this thing? Or is it too expensive?"
In his opinion, these two reasons are not big problems. If there are difficulties in production, then expand the scale. It is very expensive. It seems that this thing is very similar to the DJS-60D. The 60D can afford it. This thing In fact, it’s no problem.
So he was very curious about the reason why Gao Zhendong did not recommend using it. In short, it couldn't be because Gao Zhendong was reluctant to part with it.
Gao Zhendong shook his head: "No, as I said just now, there is no problem with self-production. It is definitely expensive, but compared with radar, it is not too expensive. Overall, it is still worth it. I mean, this The thing in my plan is just a semi-finished product and has not yet reached its final form. If you use it too much, you will have to change the design when the final form is completed. It is okay to use it a little, at least to design a prototype first. Still very appropriate.”
Everyone let out a sigh of relief. It turned out to be like this. It scared me to death.
"General Engineer Gao, you may think this thing is not perfect, but in our opinion, it is enough. We really can't afford to wait."
"Yes, yes, we think this semi-finished product is good enough to use. Let us use it first."
The comrades present at the meeting were surprised by Gao Zhendong's high demands on himself, but also strongly demanded in unison, if you want to change this thing, you can continue to change it, but for this version, get us a batch first, time waits for no one.
"Well, Mr. Gao, we must have a batch. The electronic countermeasures you once told us are so needed. We went back and studied the countermeasures you gave us carefully. They are all very good, but most of them are We can't do it. The back-end processing technology is too backward. We can only drool at the good things. Now, you have solved the biggest problem of high-speed processing devices. I don't care how bad you think this thing is. Perfect, I think it can be used, haha”
——I don’t want what you think, I want what I feel.
Electronic countermeasures have very high real-time requirements. No matter how many DJS-60Ds are piled up, they will be useless in many cases. This is not a question of whether they are expensive or not.
Unexpectedly, Chief Engineer Gao was not only responsible for providing solutions, but also quietly solving basic technical problems! The comrades engaged in electronic countermeasures looked happy.
Next, many comrades made similar remarks, Mr. Gao, you are great, but you think this thing is not good enough. We are not good enough. We think it is already very good, and we strongly demand it to be standardized and promoted.
Seeing the "intense anger" of the comrades, the leader of the Defense Working Committee looked cheerful and said with a smile: "Comrade Zhendong, the comrades' voices still need to be resolved. They are very anxious. We will decide on this matter." Produce a batch of individual models first and let comrades use them first, so as not to delay the development of technology, and secondly, let everyone accumulate experience first. Once you make it, they will just modify it.”
Everyone nodded: "Yes, yes, we guarantee that we can keep up with the improvements."
Gao Zhendong couldn't say that this thing was in his plan. In fact, it was just a chip in the end. Since everyone strongly requested it, it was not impossible.
He nodded: "In that case, let's produce a batch..."
Before he finished speaking, the leader interrupted: "Please give this thing a name. It cannot be placed in the 12th Machinery Department for the time being. It will be hung under the National Defense Work Committee. Then we will organize production here."
He wore glasses and had better eyesight, and he could already clearly see the scratches on the surface of the board.
He also realized that this thing should be built with the SZ61 series chips mentioned by Gao Zhendong. Once the circuit diagram and component list are obtained, it can be easily copied.
So he stopped the results without hesitation and followed the rules of the National Defense Work Committee. The confidentiality work should be easier than that of the 12th Machinery Department.
(End of chapter)