Lessons Learned In 4G LTE

Finding a happy medium between power and complexity is an ongoing challenge, and it’s only getting worse from here.


By Ann Steffora Mutschler

While 4G LTE has moved into the mainstream, there are lessons to be learned about these very complex modems, especially from the perspective of balancing power and performance.

The road to mainstream wasn’t exactly smooth sailing.

“4G LTE initially got a bad rap for battery life, for power consumption,” said Pete Hardee, low-power design solution marketing director at Cadence. “What’s actually going on there is that the energy per bit is really good and better than previous technologies—it’s just that we are sending a lot more bits. The data rates are a lot higher. The data rate for 4G LTE is 100 Mb per second maximum on the download and 50 Mb per second on the upload. What 4G LTE is capable of today is a lot higher than what came before.”

While that data rate and the transferring of more data is going to suck a lot of energy, he noted that when you look at it from an energy-per-bit perspective the technologies are getting more and more efficient. “What’s interesting about this is the perception that people had, especially when 4G LTE first came out, that there was a huge increase in energy usage compared with 3G. That same thing happened with 3G compared with what came before. When the network gets upgraded and they can all of a sudden get much faster access to data, they’re just not appreciating the energy that uses even if the technology is getting more efficient.”

From a quasi-technical standpoint, Chris Rowen, a Cadence fellow and CTO at Tensilica, explained what makes 4G LTE more complex:  “In order to get these high bit rates—that is, to get many bits per hertz of spectrum—if I’m going at 1 Hz I ought to be able to get one bit for every hertz that’s in the spectrum that I use. But because of sophisticated encoding and error correction and the use of antenna diversity, I’m really now expecting to get five or six or seven bits per hertz and that drive for spectral efficiency. How many bits per hertz can you communicate through a wireless channel has been something where people keep pushing the envelope in theory, and LTE actually needs to deploy those techniques in practice. It’s quite remarkable how aggressively people are taking the theory and translating it into practice.”

Turning theory into practice in spectral efficiency requires a number of advanced techniques to be applied. “Mathematical complexity translates into a couple of things,” said Rowen. “One, it takes a lot of hardware to do it. There’s a lot of computation. And, it’s quite sensitive to the particulars of noise because, after all, this is like Shannon’s Law. I’m trying to get the information through a channel in the presence of a great deal of noise and there are different kinds of noise. There are lots of real-world issues in making things reliable in cars, in urban canyons, on trains, in the presence of all the other electronic noise we have. In practice people have to work pretty hard to figure out how to make these things work reliably in the presence of all these real-world noise sources.”

For example, modern wireless systems rely on a technique generally called orthogonal frequency division multiplexing (OFDM), which means the available spectrum is being broken down into both little tiny time windows and little tiny frequency windows, in which a user’s burst of bits is allocated to get through from the base station to the phone or vice versa. Because it has been broken down into very tiny buckets of frequency, it is quite sensitive to Doppler effect, meaning when you are speeding along in your car at 80 miles an hour if you are going away from the base station, your bits are getting stretched out and your frequency is shifting. When you are driving toward the base station your frequency is getting closer together and these frequency shifts have to be taken into account in the system.

“This is one of the dozens of different kinds of things that these wireless systems need to take into account so it’s very complex from a mathematical perspective. It’s also very complex from a practical engineering perspective to do this but people are working it out and there are quite a number of successful LTE modem silicon examples,” Rowen said. This complexity has also driven new approaches in adopting more programmable approaches.

As to how the extra math relates to power management, “In simple terms it just makes it a heck of a lot harder because math takes energy. You have to come up with different architectures that are ever more closely tuned so you are doing the multiplies just when you need them and just with the number of bits of resolution that you require. And you need clever techniques in your digital signal processors to reduce the energy that is required for all the phases of, not just the multiplies itself, but moving the data around and between memories and other registers and doing the other computations,” he added.

Balancing power and complexity with 4G LTE modem design is an extension of a classic problem, said Johannes Stahl, director of product marketing for system-level solutions at Synopsys, who first starting discussing algorithms and tools with colleagues when they were still university students. “In the specific space of modems you have the algorithmic aspect of LTE, which is very complex. We actually have the first aspect happen of power optimization when you ask, Are you using the most optimized algorithm? You can’t fix it later if you spent too many operations in your algorithm, and you will spend that amount in power. So the very first step is to optimize the algorithms. And, of course, it’s part of the secret sauce of everybody implementing a modem: Are you using the best optimum algorithm to get the performance at the minimum cost?”

The way to do that is with simulation, he noted. “The interesting factor is that, as opposed to many other segments of modeling, the model of the environment for mobile is now very well-known. You can describe the environment of a mobile moving either in your hand while walking or riding in a high-speed train extremely well with a computer model. Now, when you have done that you actually solved half of the problem because the remaining half is then to describe your algorithm and simulate it and also to describe the standard. There are a variety of companies that help customers by just providing a library for the standard. (Synopsys is one such company.) Then you do statistical simulations that run for very long time.”

Rolling it out

Thomas Bollaert, vice president of application engineering at Calypto Design Systems, observed that typically there is a phased approach to production and rolling out advanced modems such as 4G LTE and LTE Advanced. The first implementations may be onto a USB dongle that gets plugged into a laptop to enable mobile connectivity. There aren’t as many power constraints or form factor constraints as when modem is integrated into a cell phone or smartphone. There aren’t battery life issues because it is being plugged into a computer. It’s a great way to test silicon, as well, and once it has been field-tested with the USB dongles it becomes safer and easier to put it in a real smartphone.

Once that happens, there are integration challenges to be faced, he said, although with a better grasp of the modem itself because it’s already been to silicon. Still, that integration can get tricky. In that first SoC implementation, chances are it will be a standalone modem IP block next to the other modems because smartphones have a lot of connectivity to support. In successive generations and revisions for cost reduction, there will be opportunities to integrate the 4G LTE with other modems, such as 3G, that share functionality in order to improve area and therefore improve power consumption.

When it comes to improving that power consumption, Bollaert explained that if there is a standalone block versus an integrated block with other radios, then power optimization will be different. “You’ll be looking at coarse-grain power saving features to begin with. For instance, being able to completely shut off the entire block when you are not using it versus shutting off only parts of the modem when you have shared features between 3G and 4G, for example, you’ll want to be using one set of features and not the other, but you are not going to shut everything on and off at the same time.” Finer-grain power saving techniques are implemented after that as design teams keep pushing down for even finer-grain opportunities, which include not just shutting on and off sub-blocks but sometimes going down to the individual flops and seeing which register, which flip-flop can truly be gated or not depending on the use mode of the block.

Given that integration in today’s design environment means as much as 70% of a design is IP that is re-used, developed internally or obtained from a third party, different power optimization strategies are employed depending on where the IP comes from.

Lawrence Loh, vice president of worldwide applications engineering at Jasper Design Automation noted, “We work with several customers that are leading in terms of the space and quite surprised to see that while there are a lot of common approaches they have some very fundamental differences in terms of the power saving techniques that they use. But when I dig into it, it wasn’t as surprising because one company basically gets most of their IPs from a third-party—they apply the low-power techniques given that restriction—and the other company has built a good majority of the IP’s by themselves so they start building in some of the requirements in the IP.”

The company that gets IPs from third parties has some ability to insert their power techniques, but it’s still less flexible than if they were to start with that in mind, he said. “Say this IP is going to be used with the 4G LTE modem, which has this power requirement, so it must have this footprint, the die must utilize this much power. Then you can do more—that’s the good part. But it’s not always good because if you try to design every IP by yourself there also are drawbacks. IP from an IP provider goes through more rigorous testing just because more customers are using it. ARM is a good example. They provide IPs to many people, and by the time the IP reaches the market it’s rock solid from an integration point of view. The risk is lower so they can be more aggressive in bringing things to market quicker. If the IP is developed mostly in-house, chances are there’s only one company (which is that company) using it so it’s very specific.”

At the end of the day, from the mathematical complexities to verification challenges and power-saving techniques to be used, 4G LTE offers some interesting lessons for future modem designs. And the challenges are only going to increase going forward.