Linearization of Upcoming High-Efficient RF Power Amplifiers, Part 2: Efficient Broadband PA Architectures

Good morning. Thanks for joining our workshop. I'm Salva Finocchiaro. I'm an assistant architect at Qorvo. And I will give you a very high-level overview of architecture and design techniques for broadband and high-efficiency PA as an introduction to the next steps, like in the workshop, which are the more important, which is kind of actually the modeling of the device, and then kind of the system-level simulation.

So we're not really got deep into the design aspect, but hopefully there's enough information. So I kind of explain a little bit what's the state of the art like when it comes to efficient PA.

My talk is relatively short. I have, like, a brief introduction about the why we need those high efficient PA, and why they need to be broadband. We'll take a look at a couple of different architectures, and do a little bit of comparison. Again, very high level, but still what it really is.

And then I'll show you one example of an existing product that we have at Qorvo, and some initial data in terms of measurement that we have done, which will then be further investigated by the Roger Schwartz team with their characterization, and then by the AMCAD team with the modeling by the MathWorks team with a system-level simulation.

OK, so why do we need high-efficiency broadband PAs? So the 5G and future system, I mean, they're kind of driven by two key factors. So system capacity, which is spectral efficiency, and system cost, which is energy efficiency. But if we go take a look at Shannon equation for the channel, we kind of see right away that the capacity is actually, it scales linearly with the bandwidth, and it scales with log 2 with SNR.

So at the end, it's like, what is that we can actually really do? So we can increase the bandwidth. And how we can do that? We can do carrier aggregation, we can move a millimeter wave frequency where we have a larger bandwidth available, or we can try to increase the number of channels. We could do MIMO, Massive MIMO. The other option is to go increase SNR, but I mean, you can try to increase the PA power, but there are emission limits. So that's not, let's say, straightforward.

And lastly, we can do small cells to try to improve-- to try to improve coverage. This one plot shows that when you look, like, into the full-frequency spectrum, then even inside, there's going to be relatively large bandwidth. And there's going to be a situation where you may want to cover across different bands. I'll show that in a second in the next slide.

So the PA, we all know, it's kind of a fundamental block in any wireless system. And the main functions of the PA is to convert DC power into RF power. And the gen efficiency, is like the RF power divided by the DC power. And it's actually not even a very good converter. In fact, I think one problem is that PA consume about more than 50% of the system DC power. So it's very important to try to improve the efficiency. On top of that, modern communication systems, they operate with complex modulation schemes, which have very high peak to average.

And one problem is that-- which can be as high as 10 dB, like with a new modulation scheme. And the one problem, which is typical of PA, especially for PA operating in class A, B, or B, is that your efficiency, if you look at these curves, go-- oops. Sorry. Your efficiency goes down quickly with backoff. So as soon as you move away, like 10 dB from your peak efficiency, you are really throwing away lots of power into it.

So we really need to improve-- we need to improve the efficiency. And there is always a trade off there. So what can you do? One thing could be that you can go and operate the PA a little bit closer to the saturation point. Then, you get more distortion, which you need to correct. So you introduce DPD techniques to try to distort and create a more linear response. We will see that you will need DPD in any case.

The next one is to try to improve the efficiency curves of backoff. So instead of just traditional class B, or AB operation, we're trying to create a second peak in efficiency at certain back off. This block under shows that at the end, the data is actually distributed around this point. So we really need to find a way to improve the efficiency here. And that's one of the main focus of this new architecture that you'll see.

The other aspect which is important is that high data rates requires wide bandwidth. So there is a need for large instantaneous bandwidth. And the carrier like the carrier 100 megahertz, but there could be more than one carrier of 100 megahertz into-- at a certain distance. And minimum requirement these days is, like, about megahertz for the PA to support.

And on top of that, you may want to operate one PA to cover different frequencies, like, for example, like all-- like in this case, I'm reporting like the N77 band-- from 3.3 to 4 gigahertz. So that's also, so there is an instantaneous bandwidth challenge, and there is a operating bandwidth challenge. So both are important. Marcus has shown this slide before. So what are the things that we can do for the PA to try to improve-- to improve the efficiency?

So we have said that, OK, the efficiency goes down quickly back off. So we need to find some techniques, some design techniques, or other architecture approach, something as system level, to try to improve that. From an architectural point of view, from a design point of view, these are the three major approach. So we can do a web form engineering, which means class of operation. So we can go and try to look at spatial matching networks, try to operate the PA inside of class A B, we try to operate in class F, or things to that extent, which is this main bucket here.

We could do load modulation. And a load modulation includes techniques like Doherty, outfacing, LMBA, which is becoming popular these days, even if it's still not very well accepted by the industry. It's more at the research level, which comes into this bucket. And then we could do a supply modulation, like envelope tracking, which requires being able to dynamically track the envelope.

So at the end, there is this desire for the really largest operation bandwidth for the PA. And the main reason is to try to reduce the system complexity area and cost. So fundamentally, I'm showing here that those are the two main things we want. So ideally, we want to create a system which has flat efficiency versus power, and then unlimited bandwidth to be able to support any possible frequency.

OK, so now, let's go take a look a little bit at a couple of these categories that we have mentioned. So if you look at load modulated PA architectures, I already mentioned, like, the three main categories like Doherty, outfacing, and LMBA. What is the main difference? So Doherty is probably the most-used technique, or architecture, these days.

In Doherty, you have two amplifiers, which actually loan modulate each other. And one amplifier, usually called the carrier, is biased in class. AB and the second amplifier, which is called the peaker, because it will come contribute like a peak power, which is typically biased in class C. And the reason for the two classes of operation is that because you want one amplifier, which will contribute to power, like in the lower power regime. And then once you reach a certain power, which supposedly be your backoff--

Sorry. OK, now you can see. Sorry. So basically, like, once you reach a certain peak power, which is typically your backoff, then you want the peaker to come into the picture. So the peaker is biased in class C and then contributes to the final power range with reasonable efficiency. We will see that this kind of approach requires DPD, but most of this architecture require DPD.

So what is the issue with symmetric Doherty? The symmetric Doherty has-- it amplified our same size. So fundamentally, there is a limit in what is the backoff that you can actually achieve. And that limit is like 6 dB if you go and do the math. So if you want to try to extend that, and when you try to go, because we talk about we may need up to 10 dB backoff, or even more. So if you try to do that, one way would be to make the Doherty asymmetric, which means that you change-- adjust the size of the two PAs according to your backoff.

So for example, if you want 8 and 1/2 dB backoff, the ratio is actually 1.5. So you can make the area of-- to the periphery of the two PA, like 1.5. And you can achieve the wider backoff. The problem with the wider backoff is that now, you-- your efficiency tend to dip in between the two peaks. It's like, this is like traditional Doherty here. Like you do like an asymmetric Doherty, you see the curve tend to dip. So how can you address that? So you can address that adding additional peaker stage.

So fundamentally, you can go into three way or multi-way Dohertys. And that kind of helps you a little bit like this. But what is the price that you pay? It's complexity, and the combiner, and everything else. So at the end, it becomes like even more narrow band. If you go take a look now at the long modulated balance amplifiers, OK, so if you go take a look like a load modulated balance amplifier, what is the main advantage? Why is everyone so excited about the LMBA?

Because in the LMBA, we replace the transmission line, invert the inverted lines of the autobus stage of the Doherty, which is current combined through the inverter lines, which is kind of limiting the frequency range with a much more brand-- much more broadband power combiner. So in a way, we kind of remove one initial frequency limitation. So the LMBA basically consists of two amplifiers, as I mentioned before, which kind of modulates in a way each other.

So basically, we have a control amplifier-- a control amplifier, which we'll control through amplitude and phase, the load seen by the balanced amplifier through the power combiner. So typical in a traditional LMBA, the balance amplifier is biased in class AB, the control amplifier is biased in class C, and the amplitude and phase control, they get the phase are adjusted so that you load modulate the balance amplifier like between the backoff and the peak.

This configuration suffered by the issue that-- because your balance amplifier is acting as the carrier, and the control amplifier is acting as the peaker in a way, it will tend to compress. So you really need like a large current amplifier to be able to provide the power, like the peak power. So it will tend to compress and you gain compression.

So basically, it requires that you actually adjust both amplitude and phase when you are in this peak regime. So one way to overcome this problem is to use what is called, like, inverted LMBA. They inverted LMBA, just to switch the role of the balance amplifier and the current amplifier. Now, the current-- the control amplifier becomes like the carrier equivalent of the Doherty. That becomes, like, your main amplifier. And then the balance amplifier becomes like the peaker amplifier.

So you switch the operation, you bias the control amplifier in class AB, you bias the balance amplifier in class C, and you fundamentally gets what looks like a pseudo Doherty kind of amplifier. What is one of the issue with this amplifier is that it's true that we are removing the inverter line, and we are combining through power combining, which is broadband. But at the end, the carrier amplifier acts as my main-- my main amplifier.

So any broadband operation that I need falls 100% on this amplifier. So this has to be broadband, high efficiency to be able to support like your requirements. And one last topology that's been shown in terms of LMBA is the orthogonal LMBA. The orthogonal LMBA has the advantage that you move the ingestion point from the outputs to the inputs. And you leave the isolation part of the output combiner oven so that it acts as reflective loads. So fundamentally, you are injecting your signal and the control signal at the input port.

You reuse, in a way, the amplifier. So you can save-- sorry. You can save one amplifier. But and you rely on the reflecting nature of the oven here to modulate like the impedance on that. So if you go through the analysis, you will see that this also acts as kind of a Doherty amplifier. One amplifier will conduct more, and the other one less, as soon as you modulate. Maybe one thing worth mentioning quickly is that when you modulate from the output port, like in a traditional amplifier, inverted amplifier, you are changing the impedance on this side of the power combiner the same way.

So they both change the same way. But when you are doing this approach, they actually change in opposite way. So that's why it will eventually work, and it will look like a Doherty. So you gain the negative of this one, which is-- you say why everyone is not using this. It's because this has the same problem of the traditional amplifier. It will gain compress. And then basically, you will need to adjust the amplitude and phase dynamically when you are in the range.

And those two architecture, in my opinion, are very good if you could use it in digital drive. So instead of trying to do a single input, single output with analog splitting, and all this kind of thing, if you could drive it digitally so you have now two dots driving into the two paths, then you can dynamically adjust the amplitude and phase. Then I'd see lots of potential for these kind of topologies.

A quick comparison between the two. So just to run Apple to Apple in a certain way. So fundamentally, like I mentioned before, like the Doherty, it's kind of well-known topology how it works. The inverter LMBA works like a Doherty. So why we are not going to use this? Because when we go and look at the performance, as I said, it showed that the broadband nature of the combiner. But then, as I mentioned before, I still need to figure it out how to go and design a broadband amplifier with very good efficiency for my carrier, which is a similar problem to what you have in the Doherty.

And what you will find out at the end is that the main difference that you can see is that the Doherty amplifier will be very good, and achieve a very good efficiency with, let's say, like a much smaller kind of aberration bandwidth. But the low-modulated balanced amplifier can probably operate on a wider bandwidth, but cannot achieve the same peak efficiency that the Doherty can achieve because of additional loss through that combiner, which will affect.

So it gets like a little bit broader, but is a little bit lower. And so at the end, like the Doherty is kind of the state of the art these days. And the LMBA is still kind of in research-- in research phase. So as an example, I put here one of our products, the QPB 3810, which is like an 8 watt, 3.4 to 3.8 gigahertz GaN Doherty PAM. A PAM is our PA module.

As you can see, the module contains three amplifiers. The two, which combine at the output are the Doherty PA stage and an input driver. And we also have integrated some bias and control. So it's a nice small and compact module. You can get the information for this device or website. Just look for the datasheet and you can download all the information from there. But to the right, I'm just showing the peak-- the peak power, and the average efficiency as a function of the input power.

And you can see it's pretty much kind of a straight line. And this is like an average efficiency. And it's around, like, 50%, I think. I cannot read from here. We have also created a reference design, which was really meant to use as a demo for these kind of applications like a 320 watt, 64T64R 3.4 to 3.8 gigahertz base station antenna. And what we put on the EVB is the QPB module driver, like circulator, LNA, for DPD, and then biasing control.

So this function is integrated on DPD. And this can be easily used for DPD characterization and modulated signal measurement. I'm not showing here, like AM to AM, AM to PM of this reference design. But what is more important is probably how this device operates in a DPD loop.

And here, what I'm showing is like a basic configuration with a IMD, RFSOC in a configuration where we have 39 dBm average Pouts and 7.5dB power. And you can see that we have tested in several different carrier configuration like 20 megahertz, 100 megahertz carrier, 2 by 100 megahertz contiguous, 2 by 100 megahertz with 100 megahertz spacing, 2 by 100 megahertz with 200 megahertz spacing.

And in all cases, we were able to achieve pretty good ACPR. So this is a plot like where it shows the two carrier, which are spaced away. OK, so what is next in the workshop? So we talk about the different architecture, we talk about the need for the linearization. Now, we want to try to give you an early way of simulating this kind of system. So fundamentally, we will do a characterization, like a measurement characterization, of the device. AMCAD will redo their own characterization and create a nonlinear model for the same device.

That non-linear model created by AMCAD will be fed to MathWorks, which will create an RF blockset. So go with that. And basically, like, we will build a system-level simulation around that to show that you can linearize, and also to show that once you have this model, you can create much more complex simulation. And Georgia will have some examples for you. So with this, thank you. And we'll give it to Florian.