6G and AI in Wireless with MATLAB

[AUDIO LOGO]

Hello, everybody. My name is Houman Zarrinkoub. I am the principal product manager for the wireless communication products here at MathWorks. And today, I am joined by my friend and colleague, Dr. Iman Abdallah.

Hi, everyone. My name is Iman Abdalla, and I am part of the application engineering group here at MathWorks.

Together, we would like to welcome you to this MATLAB Expo 2024 presentation entitled, "6G and AI for Wireless in MATLAB." Let's talk about 6G. Mobile wireless communication is standard-based, and the standards are introduced every 10 years or so. And next generation 6G chances are borrows a lot from the present 5G standard.

However, it is envisioned to provide a ubiquitous and sustainable connectivity for the wireless communication of the future. As I mentioned, between two standards like 5G and 4G is about 10 years. The last time 5G was introduced was 2018. So around 2028, '29, '30, we expect to have the 6G standard.

So to achieve all these requirements that are being discussed, what are the enabling technologies of 6G? The other technology is including new frequencies and spectra, including millimeter wave and sub-terahertz, reconfigurable intelligent surfaces, and nonterrestrial network and cell-free massive MIMO systems. We're going to go through some of them in detail.

What are we doing in MathWorks to help you with 6G exploration? We have introduced, as of release 2024a, 6G Exploration Library, a collection of MATLAB functions and apps to help you explore, model, simulate, and test 6G waveforms and technologies. It is an extension of 5G Toolbox, so we go beyond what 5G standard has to offer.

And everything in this library is open MATLAB code. You can edit every function, modify it, change it, and introduce your innovations. When you open the 6G Exploration Library, you see all those technologies I mentioned to you.

For example, integrated sensing and communication, reconfigurable intelligent surfaces, AI for 6G-- examples that I show you here for different aspects, including cell-free massive MIMO systems measuring the impact of different frequencies on the link performance, link level simulation, and so on.

Why is 6G looking for new frequencies? Because chances are, if you want to increase the capacity of your network, you have to increase the bandwidth, as well as the spectral efficiency. So if you want to increase your bandwidth, you need a center frequency around which contiguous bandwidth available.

So we have to go beyond typical cellular frequencies and look for new frequencies. Upper mid-bands around 10 gigahertz is being discussed a lot by customers as a potential frequency to be offered in 6G context. Millimeter wave was introduced as far as 5G.

We don't know if it's going to continue in 6G. And sub-terahertz is being discussed more than 70 gigahertz all the way to 300 gigahertz. Now, if you're dealing with high frequencies and high bandwidth, chances are you're doing a lot of beamforming and massive MIMO applications. And when you do beamforming, your beam width is going to be narrow.

So essentially, when you have a narrow beam width and communicating in directional way, you need to embrace this notion of ray tracing. You examine the propagation effect as ray of waves propagating. We have ray tracing, channel modeling, and lots of functionality in MATLAB. For that, you can use it to model 3D environments, indoor and outdoor.

And we have multiple methodology image method, SBR method to make it fast for the ray tracing. And we support reflection and diffraction. And we support the environment where you specify not only the coordinates but also the elevation, the buildings, the terrain. All that stuff is being handled in MATLAB. And when you are building new RF transceivers for new frequencies, you want to do end-to-end simulation, the digital RF antenna array propagation.

And you see in MATLAB and Simulink, we have the ability using multiple products that we have to allow you to explore transmission baseband using our 5G or 6G functionality, RF front end using our RF Toolbox, RF Blockset, electromagnetic propagation channel modeling and inverse operation-- and have a good sense of the end-to-end antenna to bit simulation.

How about reconfigurable intelligent surfaces? It's one of the enabling technologies of 6G. Essentially, it's talking about an array of controllable passive, low-cost reflective elements. Usually when you have a blockage, you put repeaters which are active elements.

How about if you have surfaces of buildings that are controllable and reconfigurable? meaning the electromagnetic properties can be changed, using them to essentially achieve the reflection as you require. That's RIS. So each element can be reconfigured to apply a custom phase shift.

And when you do a phase shift and coordinate them, we can cause constructive interference, therefore, a higher power at that receiver. Let me show you an example. Let's say a base station is communicating with a UE or User Equipment. So if you have a line of sight, massive MIMO is very effective. And you beamform and increase the capacity.

What if there is a blockage and you don't have line of sight? And you use these passive arrays which can be controlled and configured to do beamforming. And that way, you achieve essentially, a nice channel improvement using RIS. So what we have in MATLAB in our 6G library, we have the ability for you to perform a link level simulation with and without RIS

So if RIS is enabled, you see that the signal strength is higher. Your performance is better. And without that, you see that the signal performance suffers. We can do all of that step by step in MATLAB.

How about nonterrestrial networks? The ubiquitous connectivity requirement of 6G implies that you want to add-- you want to achieve global service coverage. So NTN, or Nonterrestrial Networks, meaning putting LEO-- Low Earth Orbit-- satellites in the sky, can be used as essentially a base station in the sky. And these fast moving LEOs can provide essentially, cellular coverage to anybody on the earth, as long as they are visible to the end user.

Now, another topic that comes up is the inter-satellite links and those hops between different satellites so you can communicate between any point on Earth using any combination of satellites. Some people are looking at coverage targets of 99% of population, reaching at least 1 megabit per second data rate, and the whole world could be covered.

So NTN is motivated like that. We have done this 5G NR NTN link level simulation-- so MATLAB examples to measure the NR NTN link performance. So you have channel models that we got from the ITU standard. We use Doppler compensation before and after transmission and using the 5G transmitter and receiver link. That way you can have a very good estimate of the performance of your NTN.

So how do you download this 6G Exploration Library? You can go to MATLAB release 2024a-- and after 24b-- and look for 6G under the Add On. And you can find the 6G Exploration Library. You can go to your favorite search engine like Google and put 6G Exploration Library. And you find the MathWorks link to that to install it.

You can go to the 5G Toolbox Documentation at mathworks.com/help/5g. You find the 6G Exploration Library there, and you can download right from there. Or you can go to the source. Look for MATLAB Central or File Exchange in MATLAB and look for 6G. And you can download and install 6G Exploration Library.

At this point, I would like to invite my colleague Iman to talk to us about application of AI for wireless communication.

Thank you, Houman. Now let's discuss AI. With the advancement in and the success of AI in multiple application areas, such as natural language processing, image processing, and also the advancement in hardware, it is natural to start thinking of its impact in the wireless comms area. So things to expect-- AI benefits such as improvement of the performance when you use data-driven approaches versus model-based approaches.

You can think of an example of devices. For example, if you're trying to characterize a device and its performance or its behavior will change with time and maybe temperature, then maybe data-driven approaches would be more beneficial to you and would give better performance. Another thing is it can reduce algorithm complexity and also implementation complexity.

And last but not least, it can facilitate joint optimization of network and device operations. An example would be like resource allocation, load balancing, and such problems. The thing that comes to mind as well is examples of optimization where the problem is complex.

And historically, you either decouple the problem and solve it suboptimally but just maybe optimal for each decoupled problem. But in a joint sense, it is not optimal. In these cases AI can really shine.

These are some of our investments in AI for wireless comms. You can see all ranging from localization, spectrum sensing, channel estimation, and device identification. You can tie these also to the benefits that I was just discussing. There are some occurrences where you can basically study trade-offs between complexity and performance, which also can be done using our tools.

Now, I have discussed all the benefits. Now how about the challenges? I kept mentioning data driven. So where do you get the data and specifically good data? Because the model is only as good as how the data that you train it on-- garbage in, garbage out. So you want to make sure that you get very good data.

The second thing is once you move into modeling, you're happy with the data you have. Which model do you use, keeping in mind that training is very expensive? It's computationally intensive.

And then once you're happy with the model and you simulate it, now can you test it in real life scenarios? All the way to deployment where you can-- we get questions like, how can I generate code for hardware devices? Now I am going to dig into each and every box of these to show you how MATLAB can help. And now, let's start with data prep.

So you can augment existing data or synthesize additional data using our Wireless Waveform Generator App. I'm going to show you a quick demo of the Wireless Waveform Generator App here. This is what it looks like. And you get this dropdown menu where you can pick from a bunch of standards such as 5G, 4G, wireless LAN, satellite radar, or custom modulation.

Now, for the purposes of this demo, I am going to stay into 5G. You can do custom uplink or downlink waveforms, as well as off-the-shelf waveforms like test models and FRCs. You can go in and configure things like the physical downlink shared channel, the PDSCH, or the control channel, the PDCCH, all with advanced configurations options as well.

Now, you can do this while you can also see the Resource Grid and a Spectrum Analyzer at the bottom. If you're happy with this configuration and you generate it, there you go. It pops up in the Spectrum Analyzer. And you can also add impairments, such as IQ imbalance, AWGN noise-- in this case, a PA phase noise frequency offset. And then you can also generate that. And where does it go afterwards?

So once you export, you can export to the workspace to a file with many of these options that you see here formats, a MATLAB script, or a Simulink Block. Now let's do the script in this case. So with one click of a button, we were able to generate around 200 lines of code with the impairments at the bottom for you to be able to tweak and change whatever you need for your convenience. So this is a very nice way to automate.

There's also the Transmitter tab where if you have a hardware device to connect to or one of the SDRs that we support, you can also transmit over-the-air data. So this is very useful to you for synthesizing data. Now, if you want to acquire live wireless data, we support a number of radios you can see here on the right, all the way from an RTL-SDR to a spectrum analyzer.

OK, so this is what we have in store for data prep. What about AI modeling? The good news is you don't need to write things from scratch. Chances are you're going to look and find the algorithm that you need-- some of them here, such as machine learning, deep learning, reinforcement learning. The menu that you see here is not exhaustive.

Also, the prebuilt models for if you want to do transfer learning where you can import a model, such as ResNet, DenseNet, GoogLeNet, have this into MATLAB and in your workspace as well. And last but not least, on the bottom right, you can see the reference examples that we have online. And I really want to stress how important they are. I use them a lot. You can go in, change the code, tweak it, and use it as a skeleton. So you don't really have to start from scratch.

The other elephant in the room, which is training-- so you might be happy to know that we can support. If you have multiple CPUs, a GPU, multiple GPUs, or a cloud service such as AWS, you can configure these or they can be supported in MATLAB. Now, we've tackled data prep and AI modeling.

Now, what about simulation and test? Of course, you can simulate, test your system in software. But you can also snip out the channel that you have simulated and test with real live impairments. Use SDRs and the same I was showing from before that are supported. And last but not least, the question of deployment.

Now, you'd be happy to know that we can support. Through our Compiler Workflow, you can deploy to enterprise systems. For example, using MATLAB Compiler, you can have a web app which you can share with a colleague, for example, that doesn't have MATLAB. What happens is it generates an executable file that can be run on another computer. And you can share your results.

The other thing to think of is the Coder Workflow where you can generate-- using our CodeGen, you can generate C or C++ code for CPUs, CUDA code for GPU, and HDL code for FPGAs. So I have walked you through this full workflow that we've typically been getting questions about. And one thing I want to go into is interoperability in the AI modeling bit.

So if you have code already in PyTorch, TensorFlow, you can make use of our model conversion between MATLAB, PyTorch and TensorFlow. And otherwise, you can do code execution, so run PyTorch and TensorFlow models in MATLAB and Simulink. Let me dig a little deeper into code execution.

So you can basically call Python from MATLAB or MATLAB from Python. And for the latter, which is calling MATLAB from Python, you just need to make sure that you install MATLAB Engine in your Python environment.

Last but not least, on our website there are many examples. And we keep adding to these so that you can get started and hit the ground running with code that's already out there. You can tweak, change what you need. It's kind of-- you have access to the whole source code so that you can be more efficient with your time.

Now, let me sum up what we've discussed. We have ready-to-use AI workflows for wireless comms available on our website. MATLAB can interoperate and exchange models with Python and other frameworks. And last but not least, you can capture over-the-air signals to both train or test your AI models. And, of course, you can generate synthesized signals as well. And with that, I will hand it over to you, Houman.

Thank you, Iman. So how can you learn more about all of this? So I invite you to visit the 6G Exploration library at mathworks.com/products/6g-exploration-library. Go to our MathWorks 6G page and go to our Wireless Solution page.

To summarize, MathWorks 6G Exploration Library can help you model, simulate, and test candidate 6G waveforms and technologies. It includes promising 6G-enabling technologies like new frequencies and waveforms, integrated sensing and communication, reflective intelligent surfaces, AI for wireless and nonterrestrial networks.

And MATLAB can also make the task of AI-based system design easier by providing tools for both training and testing data generation, continuously improving your AI models, integrating them into larger system for testing and validation, deploying them onto production code, and finally, interoperating with Python and other AI frameworks.

On behalf of Iman and myself, I would like to thank you for attending this MATLAB Expo presentation.

[AUDIO LOGO]