Die große Sensorstrecke - 06 - Wärme und Sensoren 📷 Krolop&Gerst

Mr. Schlüte, we have to get to this whole algorithm-software thing soon, which is your real stag horse, where you really, really have knowledge. But your, I don't know, knowledge makes heat here. Maybe we'll start with why heat is so negative at all. That's actually quite nice, so comfortable on the skin. Yes, I mean, cold is also cool. Hexenschuss, heat on it. Yes, heat patch. Give him mustard. Yes, the thing is that heat in our case has a very big negative effect. Every time something gets warm in our sensor, it means that we accidentally generate electrons, release, let through, which we actually don't want. So that means energy retention, heat also releases electrons. Heat can release electrons, definitely, yes, exactly. And then they swirl around and are read out and no one checks that there is no light at all. Exactly, and they accidentally land in the capacitor and are then registered as if they were triggered by a photon. We don't know where it came from. Are they uniform? So could you calculate them? So do they go on all colors evenly, or do you know that by chance? Typically, yes. Typically, these thermal effects really have the same effect on all sensors. Of course, there are slight manufacturing tolerances for sensors, variations, so that one may have a little more than the other, and the other may be really hot. Then you have such a hot pixel that somehow shines all the time, even though there is no light at all. Does the term hot pixel come from there? That's a good question. Whether it really comes from the temperature or just because it's hot, shines, I don't know. But yes, the effect is present for everyone, and then by manufacturing tolerances for some stronger and for some weaker. And the problem is, if you calculate that in general again, the voltage is gone, but you also have a smaller playing field with which you can work, right? Exactly, and you lose a bit of signal wave that is usable. Signal wave! But that's maybe just a very small part. I think it's more critical that this generating of electrons, that has a random component, a very large random component. That means it's like a noise. That means we add photons once more, once less with the different sensors, and we can't necessarily subtract them again. Then we have these production tolerances, where one sensor may have 100 electrons too much, because there was something slightly different. And at 1000, it's 100. And then it's intense, exactly. And that's kind of, how should I say, an extremely unpleasant effect. But that's an effect that's not linear with the temperature. Not? It's not linear with the temperature, but cooling helps a lot. It goes down potentially. There are rules of thumb for two different effects. One is called every 6 degrees, the signal doubles, every 8 degrees, every 12 degrees. There are different models that describe this. That means when we simply cool, cool by 10 degrees, this underground noise is significantly reduced. This underground current, the dark current, which also flows when there is no photon. Hence the name, dark current. Crazy. But there is actually a little trick that goes in the direction of what you said. We now have two effects, this dark current is on the one hand, so to speak, it fills our capacitors. Or empties them. But at the same time there is a noise on it. So very briefly, to explain this. The electrons fill the capacitor, but the voltage is reduced and therefore it empties itself. We empty capacitors. But we fill, the emptying happens through more. Backwards. On the one hand, we have a base spoon that comes in and the noise on it. We don't control the noise, noise is random. We can't predict that. But what we can do is, we can have a few switches on the sensor somewhere on the side that we don't use. Except for that. We shadow them and simply measure how much they run full, even though there is no light. And with that we can determine our medium base signal. Does every sensor have that? There are a lot of sensors. Not everyone, but a lot. Why shouldn't you have something like that? Sounds pretty stupid to me not to have it. Maybe it's an additional production step. But for you, in the software algorithm afterwards, such knowledge would be worth gold, right? Yes, of course. Everything I know about correction and don't have to guess, is worth gold. The effect is that the hum, as you said, gets smaller. It stays, but at least you have the basic height. You don't accidentally represent a dark gray, even though it was black. So a lot happens with corrections and the like. Yes. But pure heat is produced by what? I mean, if I just open the sensor and hold it in the light, it won't get warm. Yes, this small light output doesn't have a significant effect yet. Maybe that's direct sun radiation from solar cells or something. Maybe it's a little different, but here it's maybe not the significant effect yet. It's the multitude of electronics, of transistors that are on there, that have to be switched, that activate the pixels, that transmit the signals from the pixels, and so on. This multitude of switching and recharging of charges, that's what generates the heat. I imagine a sensor like a circuit board, a computer circuit board, where you take a soldering iron and stick the transistors and capacitors on. But that's not how it is. No, that's not how it is. They're baked into one compound, so to speak, and that's why they have a good thermal connection to each other. That means, if a transistor gets hot... Stupid thing. But otherwise you can't get it on the surface. And not so robust. It's the surface, it's the robustness that forces you to make it as one compound. That's a hot topic. So camera cooling helps. I remember you filmed underwater, and you barely had any oxygen. We had a GH5 in a water bag. And you noticed, it had an effect. Surprisingly little, you have to say, because the GHs, the Lumix themselves, I think they're very... If you were to ask me which camera has the least problem with heat, I would say Lumix in general, and then Nikon. But you noticed that it looked different. It wasn't an ISO 100, although we filmed with ISO 100. That was far away from ISO 100. Can you say whether you saw that the black level suddenly became gray, or did it become noisy? We didn't look at it that closely. We used the stuff anyway, because it was underwater in the Caribbean. Everything looked great. In real life, it's less relevant than you think. But we thought, which ISO values did you film with? And I think the first recordings were less bad than the last ones. Very likely. If you read out a large number of pictures, and then do this process 60 times per second, or 30 times, of course, you generate energy. But the sensor, millions of pixels on it, they're not all read out at once, that would be the global shutter, but they're read out in lines, right? So actually read out one after the other. Yes, exactly. The pixels can be addressed from the edge, so that you say, I want to read you out now. Like a coordinate system. Yes, like a coordinate system. And then it picks up the voltage from the individual pixels. And that's much cheaper than having a voltage converter and a marker for each pixel, which at the same time, with its friends next to it, the global shutter, takes over this function. So you have to imagine it like sinking a ship. 23A, B, 17. Kasparov. Kasparov, exactly. Then it's read out one after the other, and you only need a coordinate system like that and then you have the voltage values. And then it happens with an R7, R4, 60 million times. 60 million times per picture. Isn't there some kind of sluggishness involved? I mean, actually light speed, but does everything happen with light speed? Yes. Here, too, there are opposing interests. You want to become fast, and you want it to be precise. And as a human being, you know that what you do fast is usually not accurate. This also applies to all measuring technology. Really? Yes, indeed. The faster you want to design a measurement, the less accurate it will become. You have fewer measurements, more statistical fluctuations, and that's exactly the case here. I mean, I thought 7.7 volts are 7.7 volts, if we stick to the 9 volt example. No? Yes, but we are in the position that we want to count individual electrons that actually interest us. And we have to be pretty precise, stupidly. And that 60 million times, times 60 per second. If I... Maybe we won't pull the pictures out of the camera that fast. Is there anything else happening that can change the measurement? Yes, the manufacturers have to make sure that the system is irradiation-proof. I don't know, if you put your cell phone next to it, it can't, because it's connecting, induce charges. Oh, that could also happen. Yes, all that. Then the electronics developers think that the whole system is already shielded and is insensitive to radiation. But of course, that's a challenge. That's super interesting, which components suddenly come in. That's something you don't even think about. What kind of cell phones or microwaves. And suddenly, we're talking about the sensor, but suddenly the housing designer is asked. Because it's supposed to be thermally connected, maybe, so that heat can be transported. And at the same time, it's supposed to be protected from the outside. Which is stupid when you connect things. That's actually stupid. You mean mechanically, so to speak. To transport the heat away. Electrically connected. Yes, but somehow it doesn't want to have it back. Now we're back to the point that you said at the beginning. These are stupid heads. They know what they're doing. That's the point, something very personal. I've been doing this stuff for 15 years. And I think I sometimes find very little respect for the YouTuber. Respect for the people who should invent something like this. You're super fast. Why didn't they do that? I expected more. So I think that I don't know. If you say now, I have a sensor and I bring out a round with the same resolution. Then people on YouTube say, seven pictures, ten pictures, doesn't bother me at all. But from the technical component it can cause a headache without end. So these technical development steps, what happens to get the data 20% faster away from the sensor. That may be a phenomenal step in the technical development, but with us YouTubers we're like twice as fast, I would have noticed. Twice as much effort. But I really have to say, sometimes there's a lack of respect. I sometimes have the impression that there's a certain habit. That it always goes forward in big steps. And you just expect that. But you also have to recognize that at certain points you reach technical limits and you can't go any further. You have to take it as it is and work with what you have. Maybe there are nicer operating concepts at some point, but that at other points you reach technical limits. We do high frequency technology in the nanometer range and count individual electrons. Think about it. I'm still flashed by some things, but I wouldn't have expected how few electrons would be released and how few electrons would be enough for a picture. But everything you say makes perfect sense. You don't have 100% conversion of photons into electrons. Losses endlessly everywhere. Exactly. I think it's interesting to know what many don't have. When you think of a capacitor you always think of big numbers. That's why the example with the last part. We're talking about small numbers. We really count more or less individual electrons. In a raw file, if we see 1 bit, which is analog to digital converter, then it may only mean one electron. For the camera you mentioned with the 60 megapixel sensor, it can store about 60,000 electrons in its full well capacity. You're talking about the 7R4. 60,000 electrons per sensor. Maximum. And it has a 16-bit sensor. A 16-bit ATC converter. That means 16 bits are 65,000 steps. That means the 60,000 electrons are represented with 65,000 steps. That means for every electron the analog-digital converter counts one up. We count individual electrons. Tino is laughing. Okay. That's 2x16. 2x16. 56,000. 65,000. And 60,000 electrons. Is the information really there? Or is it blurred? That's totally crazy. More or less, yes. The noise of these sensors, the reading noise, the dark noise at normal exposure times and normal temperatures. That's in the range of 2 electrons or something like that. You're unsure about 2 electrons. 60,000 full. Full well capacity. How do you imagine the full well capacity? Let's talk about photography practice. To always have a correctly illuminated image. And now let's put it purely technically. I hate this term. Correctly illuminated, unequally cool image. That's my opinion, but still. Do I always try to fill the 60,000 electrons? Or do I try to go into the middle range? Where are we in the normal range? If we're talking about full well capacity, then it's what the sensor can take into account. Can the converter not count anymore? And it would not perceive higher electron counts, photon counts. It would also present them as 65,000, as high as the 16 bit converter comes in. And of course we try to stay away from the limit. So that we don't get into this clipping and that we cut something. But there is a second defect. And we just have, this is going to be an extra video, we have photon noise. It's going to be irregular. And that means that our signal does not rise as linearly as we expect. We expect a certain level, but in fact it leads to the sensor. Okay, let's finish this video. I'm looking forward to diving in. We'll continue in a moment. This ending may have been a bit extreme and also blatant. But don't worry, next week it will continue. And at this point I would like to briefly explain what we thought about it. Because we were not so thoughtless. Quite simply, Hans and I did not design this route for YouTube. We had enough problems with content and the whole topic. And we did not want to divide everything algorithmically into 10-15 minute pieces and take out the flow and the whole normality from Hans. But we simply recorded the route as a total work of art. And that's why we decided to make these blatant cuts after 15 minutes and do not take into account the content. You can watch the whole thing afterwards. Or you don't wait and go to our store and get the whole video route directly. The whole thing is a name your price product with a minimum price. This is also a bit of appreciation of our work. The whole team put a lot of work into it. The whole thing has to be done somehow. And we do everything if the comments are still in the frame everything on YouTube. But if you want to have the whole thing unfiltered and without any cuts, you are welcome to support us with a thumbs up and with the purchase in the store. Otherwise, I look forward to next week. And there it will be until then. See you next time.