Episode Transcript
[00:00:00] Speaker A: Welcome to ID the Future. I'm Andrew McDermott. Today's episode comes to us from our sister podcast, Mind Matters News, a production of the Discovery Institute's Walter Bradley center for Natural and Artificial Intelligence.
You can learn more about the show and access other episodes at mindmatters.
[00:00:19] Speaker B: AI.
Greetings and welcome to Mind Matters News. I'm your load pulling host, Robert J. Marks. You don't know what that means, but you will in a little bit.
Have you ever thought about all of the electromagnetic waves in your room where you're sitting right now?
You have invisible microwave cell phone signals in the air, some of them going through your body.
Plus the signals for broadcast television and AM radio and FM radio.
Even radiation from a GPS signal is there in the room with you waiting for you to tap in.
Radar also uses electromagnetic waves.
The only thing that changes from electromagnetic wave to electromagnetic wave is the frequency. How many times does the signal wiggle every second. Even the visible light you see is an electromagnetic wave that oscillates at trillions of cycles or wiggles per second.
The range of frequencies makes up the so called electromagnetic spectrum. And managing the electromagnetic spectrum is what our topic is going to be today.
We have some really great guests. Dr. Charles Bayless is a professor at Baylor University and director of Smart Hub, that's a 15 university research center focused on making more effective use of the wireless spectrum. He is really a world renowned spectrum expert who has given even congressional testimony on the topic.
We also have Dr. Austin Egbert, no stranger to Mind Matters news. He works with Dr. Bayless and Smart Hub. He also holds a professor title at Baylor University.
And Jonathan Swindell, who's a PhD candidate at Baylor University.
And for full disclosure, I work with these guys in my day job and it's a joy to work with them and talk with him today.
So welcome everybody.
Let's go ahead and start with with Charlie.
We talked about and wrote about, we did, we did an article for Newsmax on the spectrum crisis and why somebody should care about the spectrum crisis. If you have a cell phone, should you be concerned about the spectrum crisis? What's going on here and why is it a crisis?
[00:02:35] Speaker C: Well, the reason. And by the way, thank you for having me on the podcast, Dr. Marks. Appreciate this opportunity.
The reason we should be concerned about it is we're actually running out of frequencies to use.
So every application, a cell phone, a radar, even passive scientific systems that are used to observe the atmosphere, for weather and look at the stars, all of them have to use certain frequencies and we're having so many applications arise in wireless now that we're running out of frequencies. Just, you know, think back to when I was born in 1979. No one had personal wireless devices back then. And now I do polls in my classes. And some of these students are coming in with four or five wireless devices on them at one time. And every one of the enormous growth of wireless devices, every one of those devices has to use a frequency. So we're running out of frequencies. And really the crisis is that we haven't been able to find a way to get around this problem.
I really believe that the solution to this problem is technical innovation. We have to find ways to adaptively share the spectrum. But what a lot of people have tried to do for years is to regulate the problem away. And that just isn't going to be able to solve it.
[00:03:55] Speaker B: You mentioned a couple of interesting things that even weather forecasting requires electromagnetic signals.
I think that that's news to a lot of people. How does that work?
[00:04:07] Speaker C: Well, it's interesting. We had a project on this that we were working on a couple of years ago and I learned from some others as we were working on this project that weather radiometers actually are able to to read the water vapor in the atmosphere. Water actually emissions radiation at 23.6 GHz to 28, 3.8 GHz in that range there. And so what weather radiometers do is they actually listen or try to receive that very faint radiation that atmospheric water vapor emits in those frequencies.
Well, what was interesting about the problem we were working on is cell phones were then allocated for 5G 24 GHz and there was a lot of concern that out of band emissions from the cell phones would actually run right over the water vapor signatures and their emissions and possibly delay forecast for things like hurricanes and other types of water based natural disasters.
Oxygen is another one that is used in weather prediction. It turns out that near 50 GHz oxygen has an emission. And what's special about these emissions is they happen at a specific frequency.
So you can't just tell water, change your emission to some other frequency. It's going to emit water vapor at 23.6 to 23.8 gigahertz. And oxygen needs that band near 50 gigahertz. And so we have to build our own transmissions and receptions around these frequencies or we won't be able to sense them. It will affect our ability to predict the weather.
[00:05:44] Speaker B: So interestingly, it turns out that a lot of people Want to use the same interval of the spectrum. Why can't we just simply move one of the signals somewhere else like we do on our AM radio dial? We dial in different frequencies. Why can't we just move it up or down the spectrum to different frequencies?
[00:06:04] Speaker C: Well, that's a good question, Bob, and I think it's one that has been asked a lot the reason we can't do this. And you have to also wonder, why does everybody like low and mid band frequencies? It turns out that the frequencies from anywhere just below a gigahertz to about 4 or 5, maybe 6 gigahertz are in very high demand.
Last year in 2025, there was a big argument about the 3.1 to 3.45 GHz band. And Radar and wireless communications both wanted to use it.
Radar is a Department of War asset that uses that to be able to sense and detect. So it's obviously very important.
But many wanted to auction that band off to commercial wireless to be able to bring income. Now why do they both want that band? Why can't we just use higher frequencies for one of them? The answer is that is actually a lot of people call it beachfront property. It's very special frequency ranges because, for example, the aspect size is small enough.
So radars really like this band because essentially your antenna can be made reasonably small and have an electrical length that is commensurate with the characteristics of something you'd want to detect. Maybe it's an aircraft or a ship or whatever. You can get a really good detection image of that ship or aircraft because of the small frequency. If you go to lower frequencies, then the wavelength is longer. And so you aren't able to get as good of an image because it turns out that whatever you're trying to sense actually has not as many wavelengths.
So as you get to a lower frequency, whatever you're detecting is not as many wavelengths and you're not able to get as good of an image. The other thing is the size of transmission antennas and reception antennas is actually very reasonable. In fact, the higher you go up because the antenna size is based on the wavelength, the smaller the circuit gets to build an antenna. And you might ask, well, why don't we just keep going up, right? If detection is based on wavelength and antenna size is important, why don't we just keep going up? Well, then there's the other part of the trade off, and it turns out that we have atmospheric attenuation and also just propagation loss based on 1 over the distance squared for communications 1 over the distance to the 4th for radar. And this is. It turns out though that the loss actually is based on the wavelength.
[00:08:31] Speaker B: Just to clarify, wavelength is the same as the frequency. So if you talk about different wavelengths, you're talking about different frequencies. And the longer the wavelength is, the lower the frequency.
[00:08:43] Speaker C: That's right. So wavelength and frequency are just basically inversely proportional to one another. And as you go up in frequency, then you have trouble transmitting long distances. So this is why the mid band and the low band tend to be good compromises between the issues of being able to have small aspect ratios and good detection, be able to form good images or whatever the case may be, and also being able to make sure you can propagate the distances you need to, to be able to penetrate walls and windows. Some of these things get to be real problems. As you get above even 6 gigahertz, you start seeing some real problems with distance based attenuation and trouble getting through buildings and windows.
So everybody wants that low frequency, low, low band, mid band frequency. And so that's why the arguments are existent. And you know, I think at one point, like when 5G was coming in, the philosophy was let's just, let's just go higher in frequency.
But they quickly realized, and they tried to put 5G up in millimeter wave at around 24 gigahertz, that it wasn't going to be able to do what they could do at 3 gigahertz. So then they had to come down and start taking mid band spectrum. And I think people are starting to realize that this frequency band is going to be needed. We aren't just going to be able to go higher in frequencies. You've got to find better ways to share the low and mid band frequencies.
[00:10:04] Speaker B: So therefore you're saying that not all frequencies, not all wavelengths are created equal. We have a supply and demand sort of thing where there's lots of demand for a special interval of frequencies. Now, your proposal solution and emphasizes sharing spectrum. You mentioned that walk us through an example of for example, a radar and a wireless network sharing the same interval of spectrum.
[00:10:29] Speaker C: I think this is going to be really important for 3 GHz. And so what we are working on at Smart Hub, as you mentioned, our universities that are in our center, that we've got 15 of them and they're all working hard to help the United States maintain its spectrum capabilities. But the idea would be that a radar might be operating in that band and it senses that a wireless communication network is using the same frequency. It could Just move to a different frequency.
So essentially, it would just be like if you were walking towards someone on the street, you would shift over to move out of their way. And that's the same idea here, is that we would want agile systems that can actually change themselves in frequency, but do so without losing capability. So we would want this radar, if it were operating, say, at 3.1 GHz and a communication system started operating near it at 3.1 GHz, it would need to move to 3.4 GHz, but without losing radar range. And that's tricky because typically this radar circuitry and its system operation are all designed for 3.1 GHz. So it would need technology that can actually adapt, be able to sense the surrounding environment and be able to respond to it. And then also to reconfigure. We would want to be able to reconfigure our circuitry so we can, once we move in frequency, we can get our range to where it was at 3.1 GHz by essentially redesigning ourselves on the fly.
[00:11:53] Speaker B: So conventional electronics is kind of hardwired for a single frequency. The solution for spectrum sharing is you're going to go from different frequency to different frequency. And to do that, if you're at one frequency and you switch to another frequency, if your electronics are hardwired, you're not going to get the best efficiency. And so you have to actually reconfigure the electronics in order to make things better. And that's a whole new world of electronics.
[00:12:20] Speaker C: Isn't really is. And especially in the radar world, we haven't had the technology to be able to facilitate this type of movement because you have to be able to reconfigure quickly. We're talking on the order of milliseconds here or less.
And you also have to be able to handle high powers for either a radar system or a wireless communication base station. You need to be able to handle large amounts of power. And this combination has not been seen in reconfigurable devices until recently. And some recent breakthroughs in the technology area with reconfigurable devices and circuits are now showing this is actually feasible.
[00:12:57] Speaker B: You show that AI can help in spectrum sharing. Your group does. Before we discuss this, one of the things that you have to reconfigure is the power amplifier. What's the big deal about power amplifiers, and why is it so important to radar and wireless transmission?
[00:13:14] Speaker C: Well, you can kind of think about the power amplifier as the last element before A transmission. So in any type of wireless situation, you have a transmitter and a receiver. And so right before the transmitter, the power amplifier has to boost the power so that it can actually make it to the receiver. That's really the simple way to think about it. So large amounts of power have to be put out and you can't get something for nothing. So if you're going to put out large amounts of power, you've got to supply it with large amounts of supply power. Essentially you would want to have DC power, direct current power going to it to make sure it's kind of like a battery. Right. You'd have to actually provide battery power to this to allow it to boost the signal that you're trying to send so it's strong enough to make it to the receiver. And because of the large amounts of power that an amplifier puts out, it's actually the largest power consumer in the entire transmitter.
Most of the power in the transmitter goes into more than any other component into the power amplifier. And the power amplifier typically will be operating in a way so that it can use the battery power most efficiently. It will be operating in a way that actually has a lot of different spectral pieces that come out of it also. And so it might have tendency to interfere with other devices. So being able to negotiate trade offs in that amplifier. So how can we get as much power out as possible?
Meaning we maximize our radar range. The more power that comes out of that amplifier, the more radar or communication range is maximized. But also make sure it's not splattering over into other bands that are being used by other devices. And we need to be able to do this in real time. We need to be able to tune and make sure that as we're tuning and changing things, that they're actually helping this transmission be better in both these ways. And then output power means something else. It means power efficiency. Right. And we want to use the battery power as efficiently as possible because this is the largest consumer of power in our transmit chain.
[00:15:06] Speaker B: Let me transition over to Dr. Egbert and talk more about this. One of the ways that we, if you will, tune the power amplifier is with a technique called load pull. And usually a bunch of experiments are required in order to do load pull.
Without using techie words or math, what are you adjusting, what are you tuning? And why is the conventional load pull taking so long? Then we're going to get, after you explain, we're going to get the use of AI and actually mitigating this problem.
[00:15:39] Speaker A: Yeah. So if it's all right, I may introduce one techie word, but I promise I'll explain what it means. But that's the concept of impedance. And so to kind of explain what I mean by that is I'm sure everyone's probably had the experience of being in, for instance, a well lit home at night, and you're able to see your reflection in your windows.
That's the light inside of the home bouncing off of that window and reflecting back to where you can see it. But if you were to go outside, you'd also be able to see what's going on inside the house. The light's also traveling through that window. And so you have this mixture both of reflection and transmission through that glass. The reason for that is because the glass itself has a different impedance to the light passing through it than the air on either side of the glass. And where you have that change in impedance, that change in material property, essentially you end up with a mixture of reflection and transmission at that boundary. There's a similar type of behavior that exists with electronics as well. As you go lower frequencies much, much farther down than light is in terms of the frequency of that electromagnetic wave. And so an amplifier like Dr. Bayless was talking about will have that sort of mismatch occur as you go from one component to another component, you'll end up with reflections of the power that you're trying to generate and also transmission of that power that you're trying to generate in between these different components as you go along. And certain impedances will be better matched to the amplifier to give you your efficiency and output power and things like that. So this concept of a load pole is we're looking at the impedance that the load for the amplifier presents back to the amplifier. And in a load pole, what we do is we have special equipment where we can present lots of different impedances to the amplifier and see how it performs if it's trying to drive this type of impedance or this other impedance. And that's a really crucial design criteria for an amplifier designer, because they can choose, and they have ways of transforming the impedance through different types of circuits so that the amplifier sees the type of impedance boundary that works best for its efficiency or output power in terms of why this can be expensive or time consuming. One of the more traditional ways of trying to do this sort of load pull is with a physical mechanical tuner that's able to move different physical components inside to change the type of reflection that gets generated. And so that takes time it can take on the order of a few seconds, depending on the model of the impedance tuner is what we would call it that you're using for that load pull. Or it can be really expensive. You can get really sophisticated hardware that's able to synthetically create a fake reflection and send that back towards the device as well.
Those tend to operate a lot faster, but it's a lot more costly than sort of the more traditional mechanical style of an impedance tuner.
[00:18:38] Speaker B: So I really like your example of the window.
In a way, you're controlling parameters of the window, and you're trying to tweak those parameters so that you get, for example, minimal reflection and mostly transmission. But this is an iterative procedure. You have to dink it one way and. Well, that works a little bit, and dink it another way. And have I captured this idea correctly?
[00:19:01] Speaker A: Yeah, yeah, that's essentially it. Yep.
[00:19:03] Speaker B: Okay. Now, one of the interesting things is that these tweakings, if you will require a number of experiments. You have to tweak it a little bit. Oh, no, that didn't work. You tweak it a little bit the other way. So one of the inventions that you have initiated, which I think a patent is either being filed or in consideration, is a way to reduce the number of experiments. So you don't have to take as long with this tweaking in order to get the power amplifier to do what it's supposed to do. Now, I really think that what you did was ingenious. There was a technique in artificial intelligence where if you had an image and a part of that image was missing, you could bring it to AI and AY would try to fill in this image called image completion. So you applied this technique in order to reduce the number of experiments that were required to do load pull.
Could you unpack that a little bit?
[00:20:02] Speaker A: Yeah, yeah. So to kind of explain this a little bit, the type of image that we're talking about, there's a certain type of chart for those interested. It's called a Smith chart, I think Veritasium or somebody actually just put out a video recently on the Smith chart. I haven't watched it yet, but if anyone's interested, they can go track it down.
[00:20:17] Speaker B: I have watched it, and they put this big Smith chart on a floor, and it's a floor almost the size of a gymnasium. And as the guy is explaining the Smith chart, he walks around the Smith chart and says, something happens here. And then walks around the Smith chart and says, something happens, happens there. It's a great video.
[00:20:36] Speaker A: Yeah. And so that's the type of chart we would typically use when doing a load poll. As we measure the different impedances, those map to different points on that Smith chart. And what's really nice is if you're looking at certain performance metrics for an amplifier, like its output power or maybe its power efficiency, it forms these nice kind of ovular, sort of close to circular, not quite circular contours that have this really nice regular structure to them.
And so rather than measure all of the different points that you might be interested on the chart, I was thinking there's enough of a structure here that surely if we take just a few points, an AI model ought to be able to recognize the pattern and sort of fit the remaining contours to that partial data set. I was inspired by many, many years ago, when Photoshop first added Content Aware Fill, you could take a picture. My favorite example is a golf course. There's a little triangular par sign sitting in the grass, and you can come in and select the portion of the image with that par sign and hit delete in Content Aware fill. And it regenerates what the grass would have looked like if that sign wasn't there. And so with that experience going, if AI is able to do this on this full color, really complex image with all these blades of grass, surely our nice simple sort of concentric, ish sort of contours, surely it can learn that from very few points. And we found that in practice it was able to do that.
[00:21:59] Speaker B: That's incredible. I'm going to try to find that Veritasium video and put a link to that in the podcast notes. It's very informative.
One of the interesting things about optimization is that we talked about just one criteria of minimizing the reflection. There's other aspects too. And engineers know when you design stuff, you have to design around many parameters. For example, if you're designing a car, you got to consider mileage, you got to consider safety, you have to consider price and a number of other parameters. And when you design, you have to talk about all of these different parameters in your final design and make sure that you pay attention to them.
So the technique that Dr. Egbert came was concentrated on on optimizing around a single parameter.
Jonathan Swindell has taken that and he's trying to generalize it. And the problem with this generalization, if we have a number of different criteria, it kind of adds. If you're a mathy sort of person, it adds a new dimension to the design process every time you add a parameter. So, Jonathan, you extended a 2D performance map into a higher dimension.
Tell me why that's useful and kind of what you did.
[00:23:13] Speaker C: Yeah, sure.
[00:23:13] Speaker D: So you can think about.
Remember, we're trying to develop a technology that reduces the number of measurements that you need to understand this power amplifier's performance.
And so if you're using this load pool extrapolation technique in 2D, you still may have to take many extrapolations because there's a lot of other dimensions in this. You can kind of imagine this higher dimensional space.
You still are going to have to take a lot of measurements. And so even though you've reduced the number of measurements to get one of these charts completed, one image, there's actually many, many images in this higher dimensional space that describe the device's performance.
And one of the main things we care about, like Dr. Bayless was explaining earlier with power amplifiers, is the relationship between input power and output power. And so one of the things that I've done is take the technique that Dr. Egbert has developed and then look at ways to generalize this machine learning architecture to support three dimensions, or Smith tube. So you have basically lots of Smith charts stacked on top of each other, and as you go up or down in the cylinder, you have higher or lower input powers.
And that has the effect of now, instead of needing to take maybe seven points on your Smith chart, maybe you just need to take 16 points in this three dimensional space. And then you get not only the Smith charts completed, but you get that done over input power as well. And so this technique allows you to reduce the number of measurements that you need to take to understand an even larger part of this operating space for the device.
[00:25:09] Speaker B: Okay. And you're using Dr. Egbert's idea of what's the AI called? GANs, Generative Adversarial Networks in order to do this.
[00:25:20] Speaker D: Yeah, that's right.
[00:25:21] Speaker B: Okay, so let's now connect the laboratory results back to the spectrum crisis.
We know now that we have to switch between different frequencies. That's going to require reconfiguration and adaptability, and that each one requires like load poles and things of that sort. And back in 2020, experiments took seconds, if you will. But if we're going to do this in real time, especially in warfare, for example, we got to do this really quick. So, Dr. Bayless, could you, can you talk about how, how this tuning has improved over the last six years?
[00:26:00] Speaker C: Well, it really has, and I appreciate you bringing this up because so we started, I guess it was. We started back in. Bob, I think you and I started working on this back in around 2010. And we were.
It's been a long time. We had this. We had this idea. You know, I was informed by a colleague who worked in the military that coexistence with radars would be a big issue. And so when I started my career here at Baylor, I really wanted to figure out how we could design reconfigurable circuits that would be able to solve the problem. And so we started working just to see if we could design algorithms to do the load pull more quickly and thinking about eventually doing this load pull in real time on board a radar or on board a communication system device.
And we just use a commercial off the shelf measurement tool that many microwave test laboratories have in them.
And this is very slow. We were just trying to prove the algorithm. So it took minutes to reconfigure. And then we, as we started to look at the problem, we got some funding from the Navy and from the National Science foundation to look at the algorithms. And we demonstrated feasibility of the algorithms. And one roadblock that was presented to us was the power handling capability.
And so we started working with colleagues at Purdue University to try to address that. And we moved to a mechanically actuated evanescent mode cavity tuner. And if you don't know what that is, don't worry about it. It just basically moves these levers up and down on top of resonant cavities to change the impedance. And we were able to handle near 100 watts of power, 90 to 100 watts of power in this 3 GHz range, while being able to tune it was a little bit better than the minutes that our commercial load pull tuner. We were able to get this down into seconds. And then when we moved to, we started realizing that because we actually had to physically move something that was what was taking time, and it wouldn't allow us to get to that milliseconds order of time that we would need to be able to actually change in a real time sharing with radar and communications. So that's when we started working with Purdue on the idea of using silicon plasma. And silicon plasma is electricity, basically. It's a plasma that you can illuminate, and when you illuminate it, it turns it conductive, and when you take the light away, it turns into an insulator.
So we can use this plasma to build switches. And Purdue has demonstrated switches from this. And so we were able to build a tuner, basically that can load pole just like the commercial tuners we used, but it could operate in hundreds of microseconds. We could actually run a complete algorithm to tune this tuner. So this now. And it could also handle between 20 and about 65 watts over the 2 to 4 gigahertz range. So this now demonstrates feasibility of, hey, we could now put this into a system where it's small enough, it's miniaturized, it's fast enough, and it can handle power large enough powers that could be used in an application. So really, in about a 15 year span, our experiments went from working on something that took minutes to optimize all the way down to now under a millisecond.
[00:29:07] Speaker B: And where did it start? At a few seconds, Is that right?
[00:29:10] Speaker C: No, we were working in minutes when we were dealing with the.
[00:29:14] Speaker B: Okay, I forgot.
[00:29:16] Speaker C: It's been several orders of magnitude, Many, many orders of magnitude improvement. And how does this relate to solving the spectrum crisis? I think it really makes solving it technically feasible. That's the big question here, is we need a solution that isn't just more regulation. That's more than just Congress deciding we're going to take this out of the military's hands and give it to commercial wireless, or we're going to take it out of the commercial wireless hands and give it to the military. We need to design, we need technologies that can enable sharing. And until now we don't have some of these basic technologies that actually can make sharing possible. And so we've been forced to regulate, but now technology is emerging that can actually make real time coexistence possibility. And so that's a game changer for our nation. It's a game changer for defense, it's a game changer for scientific systems, and it's a game changer for wireless communications.
[00:30:09] Speaker B: This is great. I think that there has been a conflict between the military and private enterprise on the use of certain intervals of spectrum. In fact, there's been big auctions where the US Government, who controls legally the spectrum and what part of the spectrum can be used for what. And there's been auctions that have actually given formerly military assigned spectrum to the private sector. But we all want to be good neighbors and we do need a thriving free enterprise system. We do need a strong defense militarily. And I think with this idea of spectrum sharing, we can do both. And I think that that's the dream and that's pretty exciting. Okay, last topic. I want to ask Jonathan about this and maybe Austin also the big new thing now. And, oh, and Charlie, you know about this too. The big new thing about artificial intelligence is something called agentic AI, and it's a new tool. It's the new tool on the block, and everybody is using about it, using it. So let's go to Jonathan. What is agentic AI and why are people so excited about it?
[00:31:21] Speaker D: So you guys remember a few years ago, everyone started talking about ChatGPT, and you could go to this website and you could ask it a question and it would spit out, just quickly generate a text response. That was really good.
And I remember when I first tried that, it blew my mind. I was like, this is so cool. I can ask it to do poetry about whatever I want and it'll do it in seconds.
[00:31:48] Speaker C: That is so cool.
[00:31:49] Speaker D: Well, it turns out that that same underlying technology is also really good at things like software developing certain types of code and a whole lot of other tasks. And so the last few years, people have taken these underlying large language models and they've built basically software harnesses around them so that they can understand what's on your computer, what's in files, and, and do different tasks. You're basically giving it kind of this, this ability to interact with other stuff in your computer. And there's a lot of really cool uses for that. And I think right now a lot of people are, you know, there's a lot of hype, there's a lot of fear, I think. But I think it's kind of just like the Internet in that, you know, when Wikipedia came out, people were really worried about, you know, well, our students just going to go plagiarize off of Wikipedia. And of course you could do that, and you should not do that. But Wikipedia is also really useful, and I think agentic AI is super useful. And everyone's, you know, got strong feelings about it and is trying to figure out how to use it in ways that are helpful.
[00:33:10] Speaker B: Okay, so how is agentic AI being used in this Spectrum crisis, this Spectrum Shar sort of, sort of phenomenon? Charlie, can you address that?
[00:33:23] Speaker C: You know, Bob, it's really interesting you bring these two topics together, AI and the Spectrum crisis, because really, they haven't been yet connected in the public policy realm.
And we know there's a lot of push towards AI right now, even from the federal government. We know that there's been a lot of discussion about Spectrum at the federal government level, but what we really need to be doing is using AI to solve the Spectrum crisis. And I think the way we can do that is simply by automating how systems perform and interact with one another in the use of the spectrum. This is a turnkey solution. It really is. It's something that we can use AI to predict spectral use. We can use it to observe past spectral use. We can use it to be able to tell us what bands are going to be available for our use so that we can make better decisions. It's not going to be a magic genie by any, by any stretch of the imagination, because we know AI has limits and we know that users change their behavior and all these types of things, but it can get us a long way down the road. The other thing it can do is it can actually help us with these circuit optimizations and these system optimizations and some of the work that Jonathan's been doing and that Dr. Egbert's been doing, that you've heard about already today, could be transported into real time.
And this allows us to actually cut through multiple dimensions of real time optimizations to be able to get that optimization done more quickly so that we can more quickly move our spectral use to a different band and then be able to optimize our performance.
And, you know, I think it also, you know, so, so that's one application. I think it's a huge one. And I wanted to make this point on this podcast that AI and spectrum policy need to talk to one another, they need to come together. And at the federal level, there needs to be a concerted effort to bring these two things together in solving the spectrum problem. I think the other thing that we want to talk about and that I think is useful, is the way AI is being used to design circuits.
And this is a big topic right now in my technical area, which is the microwave circuit design area for high frequency circuits, everybody's talking about how to use AI to better design circuits. And software design companies are now coming out with ways to actually embed your own AI code right onto a platform where it can be easily used. And we've, we've been working with a lot of these tools and even putting some of our own developments in here. And this is going to be a powerful way to take sage designers advice. Where some sage designer who's been, who's had a 40 year career in designing amplifiers would come in and be able to design an amplifier quickly. And you say, how did you do that? Well, our goal is to be able to take some of that sage wisdom and automate it so that younger, less experienced designers can come in and get the same quality of design. So we'll hopefully have Better circuits that we can actually deploy. So that's kind of the two pronged way we can attack the spectrum crisis. The first one is doing the real time spectrum allocation and circuit and system optimization decisions. And the second way is we are just going to be able to design better circuits from the get go because we'll be able to have this tool available to us as designers.
[00:36:28] Speaker B: One of the interesting things is agentic AI is going to impact people who have spent their careers optimizing electronics.
We hear about AI replacing jobs. Do we see this agentic AI replacing jobs of engineer designers?
[00:36:46] Speaker C: My answer to this would be, absolutely not.
What we're going. There was probably a similar rumble when the calculator came out. They probably said, oh, we aren't going to need people to use slide rules anymore. And so everybody who knows how to use a slide rule is going to lose their job. And then the next thing probably happened when Computer Aided Design software came out, people thought, oh, we won't have people needing to take pieces of copper tape onto an amplifier to tune it in the garage anymore. And so all these people are going to lose their jobs. Well, they didn't lose their job. They just came to use and steward the new tools with wisdom and expertise.
And that's what's going to happen here with this AI revolution is we are going to need people to properly steward AI to improve the design workflow.
And these are going to be people who know the fundamentals of design. They know basic circuit theory. I'll give you an example of this. We were sitting in a research meeting the other day and we had been trying to use AI to design a circuit, and it came up with the circuit that basically it said it was going to be able to bias a transistor, but it short circuited the bias straight to ground by putting two inductors in series with it.
And this is to ground. And it had been trained on pictures of circuit design, but it didn't know the fundamentals. And to me, that was a perfect illustration of how when we have a real engineer who knows the fundamentals, that walks into the room, they can say, this is a mistake. Let's guide AI to make, to correct itself, to move a different direction.
AI will still make the design process in the design workflow much more efficient, but it will need to be stewarded by someone who knows the fundamentals. And so as a teacher, my job in the classroom is to continue to make sure that my students coming out, number one, know the fundamentals, and number two, know how to use modern tools to be able to implement those fundamentals in a, in a very efficient design workflow. That's the future. I don't think, I don't think you're going to see electrical engineers designing wireless systems lose their jobs anytime soon.
[00:38:45] Speaker B: Okay, that's very encouraging and I think goes against the idea that we hear all the time that all these jobs are going to be eliminated. In fact, I think that AI, as you said, is going to enhance our ability to do things and like all technology in the past, is going to make our lives a lot better.
So let me end up and ask you to put on your forecasting hat be futurist and kind of tell me where you think this conversation would be in say five or ten years from now. What, what do you think is going to happen or should we even go to that forecasting? Somebody said that forecasting is dangerous, especially if it's about the future. So I'm sure that we're not 100%, but I'm wondering if anybody has any ideas on that to kind of wrap us up here.
[00:39:35] Speaker A: So I do think, given the current trends in spectrum management at the moment, that we are trending more towards having some type of real time management of the spectrum in sort of shared bands. There have been developments over the last about decade and a half, two decades or so in opening up these sort of bands used as a testing ground almost to allow coexistence of different systems. We have the Citizens Broadband Radio service where there's cloud based systems that take in requests from unlicensed users to try and coexist with some military systems. We have in the 6 GHz band with modern WI fi routers and other unlicensed uses coexisting with some fixed point to point links like you'll see on towers for cell service and things like that, where it's starting to become a more common thing. And I think as these techniques start to have more benefits and are proven out more over time, there will be more of a shift in that direction to have this type of real time management of the spectrum. I could be wrong. There's a lot of other reasons why people may not want to go that way.
Sharing has additional complexities over just fixed allocations, but it can be a lot more efficient in the moment to moment timeframe. And I think that'll have some appeal, especially as we get more and more devices trying to use essentially the same amount of spectrum.
[00:41:05] Speaker C: Austin, let me just piggyback on your comment here. I think that one of the so Technology is one of the things that needs to be able to be brought along to facilitate this. And we're seeing maturity come to pass. And so I think you're right. As the technology continues to mature, we're going to see this. There's one other thing that's come to my attention that's really of interest and it's probably heavily because we have Bill Lear as part of Smart Hub and he's a well renowned spectrum economist.
And I think that the other hold up right now is money.
And it's not just money to develop systems, but when you think about how much money spectrum auctions net for the federal government and also wireless communications companies want exclusive use to the spectrum right now. And so they're willing to pay large amounts of money for it.
I really think what we're going to need to see is an an actual spectrum leasing scenario that is financially profitable, both for the owner of the spectrum, that is the federal government or the commercial wireless users. If they're leasing out spectrum. We really need a good leasing model, a financial leasing model. And right now we haven't really built an agreeable upon one. And so I think these deployments we've seen of shared spectrum already that you mentioned, Austin, the CBRS and the 6 GHz, I think that these would be much bigger if they had some type of leasing model associated with them because the major network operators really haven't gotten involved in those spaces yet. It's mostly private wireless companies that want access to spectrum, but the major network operators have these huge chunks of exclusive use spectrum. Why would they want to come have a chance of someone's call going down in the middle of the call? So a secure leasing model is going to be needed to push this forward. And I think that's where a lot of our research efforts need to focus. Just as they're focusing on the technology, we've got to focus on a feasible leasing model.
[00:43:02] Speaker B: So we got all sorts of problems and engineers always come to give solutions to those problems. And that's exciting and one of the reasons I love being an engineer.
We have been talking to Dr. Charles Bayless, Dr. Austin Egbert and Jonathan Swindell about application of AI to advanced electronic design and solving the spectrum crisis, which is very real. And this has been a fun, informative conversation. Thank you gentlemen. So until next time on Mind Matters News. Be of good cheer.
[00:43:44] Speaker C: This has been Mind Matters News with your host Robert J. Marks. Explore more at MindMatters AI. That's MindMatters AI.
Mind Matters News. Is directed and edited by Austin Egbert. The opinions expressed on this program are solely those of the speakers.
Mind Matters News is produced and copyrighted
[00:44:11] Speaker B: by the Walter Bradley center for Natural
[00:44:14] Speaker C: and Artificial Intelligence at Discovery Institute.
