Transforming Wireless System Design with MATLAB and NI

[AUDIO LOGO]

Hi, my name is Robin Getz and I'm an Engineering Development Manager at MathWorks. And I'm here with Jeremy Twaits, a Solutions Marketing Engineer at NI. And we're going to be talking about transforming wireless system design with MATLAB and NI. If you do have any questions after the talk or during the talk, you can definitely reach out on social media or share the Expo experience, the MATLAB Expo.

And we're going to talk about how to use MATLAB and NI to optimize your wireless systems and try and improve your product quality, get your products out faster, and lower your development efforts. By doing that, we're going to try and show you some new features and updates from both platforms that will help you to achieve your design goals and tackle common wireless design challenges and show you some innovative solutions on some new topics. Those topics are going to be kind of like six major areas where we talk about wireless standards. And those mean kind of like 5G and Wi-Fi, LTE, satellite communications, Bluetooth, AI for wireless, where we're looking at deep learning and machine learning, digital RF and antenna design was a variety of trying to do end-to-end wireless communication systems, hardware design prototyping and testing, where we test algorithms and designs over the air with either RF instruments or SDRs.

Look at some radar applications where we do like automotive surveillance, SAR, and some hands-on learning, meaning like either self-learning or direct education-type things for academia. Spectrum is a very valued resource which has tremendous economic benefit. We need to look at these things, not only from a coexistence, meaning can we implement new features and new standards and not affect the other people that are around us, but also look at new features that will enable new applications with potentially new frequencies and new products that are going to drive more economic growth.

And the way that we do this is we look at MATLAB as a common platform for wireless development, looking at three major things, mobile connectivity standards, unified design and simulation, prototyping and testing. And these kinds of things are what we talked about before in terms of standards that are standard-compliant, getting 5G signals, getting Wi-Fi signals, having an environment that's unified design and simulation, so we can design our baseband PHYs, look at our RF front-ends, look at antennas and MIMO, look at channel propagation, as well as drive these signals over the air and test them with RF instruments and SDR, potentially running some simulations on the cloud, and even generating HDL and C and C++.

Now talking more about those RF instruments and SDR, I'm going to pass it over to Jeremy, who will describe what an SDR is.

Thanks, Robin. So, yeah, I'm sure many of you are familiar with Software-Defined Radios. But just in case, a Software-Defined Radio or SDR is a wireless device that uses a configurable RF front-end along with an FPGA or a programmable system on-chip for programming on board. They can typically be used in a couple of different ways, one of which is actually tethered to a host.

So from the software defined radio to a PC or a laptop that's running a programming environment such as MATLAB, and then communicating over Ethernet, USB, or PCI Express, for example, from the host to the Software-Defined Radio. Another of another option for this is to actually operate the radio headlessly. In this sense you're able to actually push the signal processing onto the radio itself, run it on the FPGA or the RF system on-chip, for example, for things like deployed signals intelligence or signal classification.

And in one of the examples here is with the NI, an Ettus Research E-series devices, where MathWorks actually supports the ability to target the onboard processing on those devices. You can actually see those E-series modules down in the bottom left of this slide, which shows really the breadth of different types of radios and RF instruments that are available from NI and an NI brand, Ettus Research. Up in the top left, you'll see low cost, low size rates and power, Software-Defined Radios or USRPs, which are really well-suited to hobbyist or academic uses, going all the way through to really high frequency and wide bandwidth RF instruments, which are calibratable and typically used in situations where you're actually testing the wireless devices.

Looking down in the bottom row, second from right is the NI Ettus USRP X440. This is our most recent release and uses a direct sampling architecture, up to 4 gigahertz sensor frequency, but is really well-suited to integrating with external RF front-ends for being able to use in radar or satcom applications at higher frequencies, and has a pretty wide bandwidth for the realm of SDR at 1.6 gigahertz. We spend a bit more of the time within this presentation talking about the radio just to the left of that, the NI Ettus USRP X410. This is a really versatile radio that's very well suited to wireless comms applications.

Going into a bit more detail on that, we can tune that radio at up to 8 gigahertz, meaning that you can hit most of the common wireless comms bands, including getting up to the areas where Wi-Fi 6 operates. And its instantaneous bandwidth of 400 megahertz is typically sufficient for wireless communications. On board it has a Xilinx Zynq UltraScale plus RF system on chip. And you'll see that from a software support perspective, MathWorks software like MATLAB and Wireless Testbench are supported as well as the hardware support packages that the MathWorks provides.

With this software support, MathWorks also supports arbitrary sampling rate selection, through on-board IP for fractional direct digital down-conversion and digital up-conversion, which is a really unique feature that's offered through support through the MathWorks toolchain. On the next slide, Robin will give you an example of how you can actually connect Wireless Testbench to the X410.

Yeah, so here we have actually MATLAB running. We connected to an X410 and we create like a baseband object. We then set up the antennas, set up the data type as a double, so that we can send that to our plotting and utilities later. We load up a synthesized 5G 250 megasample per second map file. We set the transmit frequency.

We look at the data in terms of it just being the synthetic data, looking at the waterfall plots and the power spectrum pieces. We can see it there. And again that's the synthetic data that we just created in Matlab. We then start transmitting that over the radio. And so it downloads the file to the radio and starts continually repeating that sample file over and over.

And we just randomly capture 20 milliseconds and then start plotting that to see what our received waveform is like. We can see there's other things on the other edge that from our 5G signal, because of the other pieces at 2.4 gigahertz in the office I was at. Besides just doing comms, there's also device measurement pieces that people want to look at, in terms of what a technology known as digital pre-distortion.

So as you drive your power amplifier into saturation, it goes into compression. It has memory effects, and we can compensate for that by pre-distorting the signal that goes through. So what comes out into the antenna looks much better, or looks like a signal that doesn't have that compression or memory effects associated with it.

Now in past generations of power amplifiers, this was a pretty well-understood phenomenon. It was pretty easy to model. In next generation power amplifiers, it's beginning to be harder and harder. And when we do these kinds of things, what I mean is that it's no longer just an easy memory polynomial that you can extract with MATLAB. It varies too much with voltage and with time and with frequency.

So trying to fit these in, get your data and fit them in, whether it's captured from instruments or from models, is just getting a little bit more difficult than it was before. So we do have lots of ways to kind of look at these things, whether it's from curve fitting or whether it's model coefficients, so that you can do your envelope tracking and envelope simulation looking at your PA data, seeing how it's going. But collecting that data can sometimes be a big challenge.

And static modeling isn't enough for 5G systems with the newer generation GaN amplifiers that people are using. Circuit envelope, people want to try and do for fast RF, so that they can save as much power in that power amplifier as they can, and implement their DPD, which begins to be a big challenge for people. So capturing that data is super-critical to actually getting a good control algorithm to correct for these things. And Jeremy will talk a little bit about with the complexities and the solutions for there.

Yeah, so actually moving from modeling that DPD algorithm to being able to see how well the PA will actually perform, in practice, with that algorithm, requires moving into real hardware. And so there are multiple instruments that are required for actually performing these types of test. Not only do you need something to generate and acquire your RF stimulus and response from the device, so for that, either a vector signal transceiver or vector signal generators and analyzers.

You also require the ability to control the device under test, to put it into certain modes, for example, and test how it's going to perform under varying conditions. And so high speed digital devices are required for interfacing with the device under test in that way. In addition to digital pre-distortion, many of these amplifiers would also use envelope tracking as another means of trying to operate in a more efficient manner.

And so to apply the power, to apply the required power envelope to the PA, a precision source and measure unit is required to do that. So across this suite of instruments, it's actually possible to control them from within MATLAB using the instrument control toolbox to actually interface with the NI PXI instrumentation and to actually characterize the PA using-- but, again, both NI and MathWorks in this way.

So we have a lot of customers that have kind of gone through these kinds of problems and looked at MATLAB and hardware as solutions and actually shared their stories and shared their successes with everyone else. You can find those both on the MathWorks website and on the NI website. And these are just two of these kind of stories where both Qualcomm uses MATLAB to develop their 5G pieces, and then NanoSemiconductor also develops like system-level algorithms for predistortion by implementing things like this.

So going to the next topic of radar systems, radar systems are an incredibly complex different thing to think about from just a practical comm solution, where you have to worry more about your environment, your scenes, your scenarios as well as your radar processing system, meaning your antenna, your signal processing, your data processing, and all the resource management and controls. As we put all these things together, we try and think of it from a model-based design standpoint, where you're analyzing, simulating design, deploying, integrating, and testing.

And as you do this kind of deploy, integrate, test, you've got to get things over the air. You have to understand how it actually works in a real channel. You have to try out, potentially even simulate impaired channels so you can see how those work in the real world with your receiver, looking at a variety of different things in your radar system both with synthetic and simulated data as well as real world over-the-air with hardware data.

But we kind of go through this concept exploration, system engineering, design and test, operations and planning, and data analysis kind of phases. And we think of those in the four different major areas that radar is used in right now, where we look at radar budget analysis and look at what kind of hardware do we need, based on the budget that we have and the range that we have? What SNR do we need from our RF front end? What power do we need from our RF front end?

These kinds of things can help drive the requirements from your SDR or radar system. You know, simulating cluttering returns, looking at land and sea surface models, looking at radar reflectivity models, looking at radar surface return models, not only in peer simulation, but also over the air. We can take our transmitter on purpose, modify or impair those to emulate these models and then send that emulated system over the air to get our actual receiver, to see how our receiver algorithm and our receiver hardware is actually working.

And radar applications, so we can look at a variety of different pieces both in simulation and on hardware, where we'd want to close the signal processing loop. We want to ensure that everything is going to be working properly, that we can handle interference mitigation, we can do target detection, we can do our waveform scheduling, again both in simulation and on hardware. And then look at AI for radar, where we look at a variety of classification and deep learning kinds of pieces, where if we train our models with only synthetic simulated data, then they only work with synthetic simulated data.

We have to train our models with as much real-time, or as much real world over-the-air signals as we can, so that the models that we're using in deployment are as robust as possible. And to talk about this, I'll pass it back to Jeremy for a bit.

Yes, so you're probably getting the picture by now. At the real heart of the collaboration between MathWorks and NI is this ability to be able to combine software simulation with moving into a testbed built with hardware. And so this is another area that we've collaborated on to allow access from MATLAB to instruments like vector signal transceivers, as you see on the right-hand side.

And in this case, we've taken the RFSA and RFSG, so RF Signal Analyzer, RF Signal Generator, drivers for the VST, and we've wrapped them and compiled them into a DLL that can be called from assist from system objects in MATLAB. So we've called them here the NI.receiver, and NI.transmitter. This way, you can take some of those examples of MATLAB code that Robin showed in the previous slides, and as well as running them in software, you can actually see how those algorithms, how those architectures are going to behave in hardware.

On the next slide, you'll see the process that's followed, which is to use the MATLAB phased array toolbox to generate the desired waveform. We instantiate the transmitter and the receiver with the desired parameters, and then we can transmit our waveform and then receive that waveform back. Now, in between that, we could either have antennas transmitting over the air and picking up real targets that are in our environment, or being flown over an open-air range if we really wanted to.

Or we could even take a second vector signal transceiver, run radar target generation personality on that, to actually do some of what Robin mentioned of clutter, of emulating clutter, of emulating the environment, and also, of course, adding the targets and the return signals from those targets. Ultimately then, we can use MATLAB phased array toolbox again to plot the range Doppler map based on those acquired signals and see whether the radar algorithm is performing as we'd expect in the real world. An example-- and so just a little more detail on the vector signal transceiver.

So the latest generation goes up to 23 gigahertz in terms of its sensor frequency. This is really handy for many radar applications, because there's no need for up and down conversion for a lot of the bands within which radars are operating. And the fact that we can acquire and generate 2 gigahertz of instantaneous bandwidth is also very important for radar applications.

If you're also lurking at multichannel face coherence across an array of sensors, then the ability to share local oscillators and reference clocks across multiple instruments for face coherence is also really important, for multichannel applications. And so one customer who's put this into use, on the next slide, is the Warsaw University of Technology. And so they actually used MATLAB software with NI USRPs and PXI instrumentation for prototyping and assessing the capabilities of active and passive SAR and ISAR radar.

They utilized the IP within MATLAB for getting up and running more quickly, as well as the commercial off-the-shelf hardware from NI. And so you can see on the side of the light aircraft in the bottom right, the SAR radar prototype that they put together, as well as just above that some of the results that they were able to achieve and visualize of some ships that they were able to pick up using the radar system, so that they can actually see how well that radar prototype would perform in reality.

And so that brings us to our final piece, which is like the outreach that both MathWorks and NI does for electrical and computer engineering. We talked a lot about communications. And we definitely do those kinds of things. But there's a lot more to electrical computer engineering than just the wireless pieces.

But there are over 2,200 universities that have full campus licenses. So every student in those locations, the 650,000 students in those universities, can have full access to MATLAB and all the pieces that we're talking about, including the wireless pieces that we work with with Jeremy and the other folks from NI on, including the B200 and B210 radios.

Yeah, and so as we mentioned earlier, these radios are low cost, low size, weight, and power, can be as small as a credit card, with one or two transmitters or receivers, depending on if it's the B200 or the B210 radios, and are really well-suited for academic or hobbyist type of applications and for teaching, of course. And so just some of the feedback that we've received from the universities who've been putting USRPs into action in their courses, so the University of Southampton in the UK fed back to us that more than four out of five of their students who used USRP in their course would like to make future use of the USRP, either in their studies or their career.

And a student from Rutgers University commented that the use of USRP in practical lab assignments gave them a much deeper understanding of communication systems. So it's great to get this kind of feedback on the use of the Software-Defined Radios in academia.

And we have a lot of content on both websites, both on the MathWorks website and the NI website, for self-learners who want to learn more about these kinds of things, whether they be students or practicing engineers.

That wraps things up where we talked about standards and AI and digital and RF and prototyping, testing over the air, radar applications, and hands-on learning. Hopefully you got a smattering of different topics, that different innovations that we talked about. I appreciate everybody's time.

Definitely if you want to reach out to us during the rest of Expo, definitely feel free to do that. You can find us in a variety of different systems.

Thanks, everybody.

Thanks very much.

[AUDIO LOGO]