What you’ll learn:
- Antenna coupling within an array can affect error vector magnitude.
- EM analysis can mitigate coupling by assessing impedance mismatches.
- Look for AI as an increasingly key element of solutions to antenna challenges.
As the industry undergoes its transition to the millimeter-wave (mmWave) bands, designers must come to terms with the requirements and challenges of mmWave systems. The transition brings system architects into a world of wider bandwidths with greater data throughput, but with that comes magnified importance of aspects such as linearity and frequency flatness.
The first part of this three-video series covered the topics of frequency dispersion and interfering signals and their impact on wideband receiver performance. In Part 2, Giorgia Zucchelli, product manager for RF and mixed signal at MathWorks, discusses topics such as antenna array design, including element count, type, and spacing, as well as the assessment and minimization of grating lobes and antenna coupling.
It also explores testing beamforming algorithms and tackles issues like impedance mismatches and beam squinting in wideband systems, emphasizing the development of equalization and phase-correction algorithms.
How to Determine Antenna Array Size?
An early step in configuring a mmWave-capable system is determining how many elements comprise its antenna array. How does one choose the antenna type and the element spacing, and how do you assess and minimize the impact of grating lobes?
There are several competing requirements to consider with the antenna and the array design. For one, what's the targeted directivity and gain? The answers to this question point you toward how much power to transmit. A larger array means greater directivity, but it also increases the array’s overall real estate.
At sub-6-GHz frequencies, longer wavelengths mean larger antennas, making it difficult to manage the array’s form factor. With mmWave frequencies, more antenna elements can be packed into a given space, but it also brings more channel losses. Center frequency and bandwidth have an impact as well.
Typically, array designs try to optimize gain for the given real estate. But packing elements together will result in strong coupling. Distancing elements from each other reduces that coupling. However, it will not only result in a larger overall array size, but also larger grating lobes.
At the end of the day, it is an exercise in optimization that can be an expensive proposition. These tradeoffs are difficult to compensate for or reduce with digital signal processing. Iterations in simulation call for increasingly constrained magnetic analysis. Consequently, artificial-intelligence (AI) methods are emerging as a useful resource in this area of antenna design.
How Antenna Coupling Affects Error Vector Magnitude
It’s harder than one might think to determine whether antenna coupling is impacting system performance for metrics like error vector magnitude (EVM). Antenna coupling between adjacent elements can be likened to leakage that reduces signal-to-noise ratio (SNR). In essence, it manifests as if it were a rise in the noise floor, or a fuzzier-looking constellation.
It’s particularly important that this effect be analyzed at the receiver, which very much depends on algorithms for direction-of-arrival estimation. These algorithms assume that the information received by each antenna element is uncorrelated. So, they use the information’s diversity to estimate the location of the transmitter. Antenna coupling upsets this apple cart in that the information from element to element is no longer uncorrelated. For example, it can make the SNR appear to be lower than it is.
In massive-MIMO (mMIMO) systems, this becomes highly relevant because the beam direction is extremely narrow. Thus, if the direction-of-arrival estimation is even slightly off, the array could entirely miss the proper beam direction and the system simply doesn’t function properly.
It’s a difficult problem, because looking at the array’s SNR will not definitely address whether direction-of-arrival estimates and beamsteering will work as expected.
Assessing and Miminizing Antenna Coupling
A good starting point in assessing and minimizing antenna coupling is electromagnetic (EM) analysis. Full-wave electromagnetic analysis has two useful means of estimating the impact of coupling. The first is S-parameters, which provide insight into element impedances as well as how the elements are coupled. One can think of S-parameters as near-field interactions, or what happens on an electrical basis.
The second way in which EM analysis aids in estimating coupling impact involves what’s termed the embedded-element pattern. An antenna in free space, isolated from everything else, will have a certain gain pattern. But if you place that antenna near other antennas, the others will serve as parasitic elements that will change your antenna’s gain pattern. So, when you develop a beamsteering algorithm, you must know the pattern of each of the array’s embedded elements.
Thus, EM analysis helps with estimating each antenna element’s S-parameters and its embedded-element patterns. In terms of minimizing coupling, this information can help optimize the elements’ impedance matching. If elements exhibit impedance mismatches, it’s like a transmitter scenario in which some of the transmitted signal will be reflected. Reducing impedance mismatches will decrease each element’s contribution to coupling.
Digital signal processing (DSP) is another way to reduce antenna coupling. If you know the mutual coupling matrix of an antenna array, you can then invert that matrix as one would to achieve channel quantization. Inverting that matrix essentially compensates for the coupling in the DSP domain. However, this can be a costly proposition, especially for very large antenna arrays.
The issue of antenna coupling is a perennial one that will arise as we move to higher frequencies and larger bandwidths. It’s a problem that exists over the entire operating frequency range of the array.
The Role of AI in Antenna Design
AI is a near-ubiquitous technology these days, and it’s beginning to deliver on the promise of being a practical technology. It can be a significant factor in the move into mmWave frequencies.
One application of AI in antenna design involves the use of a surrogate optimization method. EM analysis, regardless of what software or the technology behind a solver, is a costly endeavor. The surrogate optimization method employing AI is a means of building a model on the results of fewer EM-analysis iterations. You can then use the resulting surrogate model to optimize the system.
Another way to utilize AI is in signal classification for interference, for example, or for cognitive radios. At IMS last year, MathWorks demonstrated the use of AI to analyze a signal’s spectrogram to determine whether the signal was a 5G signal or a radar signal. Surprisingly, the AI network was even able to determine which constellation was used during the transmission.
Digital predistortion presents another AI application in the mmWave spectrum. Using AI for linearization, which is a large component of digital predistortion, is akin to equalization on steroids.
These are areas being helped by AI. As AI moves more toward deep, regression learning, such new techniques will be embedded more and more to combat the increasing complexity of the problems.
Testing Beamforming Algorithms
As discussed in Part 1 of this series, one way to mitigate interfering signals is with beamforming. Typically, testing of beamforming algorithms is done in an anechoic chamber with direct measurements of the beam pattern. It’s expensive and not within reach of every design team.
Thus, modeling and simulation becomes the most useful tool for algorithm testing. This can be true even when one has access to an anechoic chamber, because the simulation results may provide insights before ever stepping into the test lab.
An important aspect of modeling and simulation in this use case is visualization of the beam patterns and using that real-world excitation of the antenna array. In simulation, the signal can include all of the impairments such as noise, the effects of linearity, and the effects of quantization of the analog-to-digital converter (ADC) as well as the beamforming algorithm itself.
Simulation enables use of a model that encompasses all impairments and employs actual excitation of the antenna using S-parameters. The model can also passively apply the embedded-element patterns throughout the entire frequency range over which the impinging signal excites the antenna.
The result is a virtual platform of your actual system that will show how the beamforming algorithm really works. Adding the insights you get into SNR, EVM, and the constellation ultimately provides metrics on how well the system is working.
AI in Beamforming Design and Deployment
To close the loop, AI is also applicable to the design and/or deployment of beamforming because it can help with the channel-inversion problem. That’s important because one must estimate the channel to properly steer the beam and implement the direction of arrival. In this space, ray tracing and channel modeling become critical, especially at mmWave frequencies.
AI can help decide how to steer the beam, considering all of the impairments mentioned earlier. Reinforcement learning is likely to be the next frontier for the design of more robust beamforming algorithms that are more resistant to different types of channel impairment.
Assessing Z-Match Losses and Beam Squinting
Wideband systems will invariably suffer from some degree of impedance mismatches and beam squinting. Assessing these phenomena begins with electromagnetic-interference (EMI) and EM analysis.
However, beam squinting is frequency-dependent behavior of the antenna. It can be analyzed in a static manner, but then must be compensated with static calibration. Though straightforward when you look at the single frequency point, it's less so when you operate over a large bandwidth of signals.
In other words, everything needs to become frequency-dependent. This isn't completely true in that you can perform static calibration. However, the system isn't static because it reacts over time as the beamforming and beamsteering algorithms continuously move and scan the region of interest.
These phenomena can be estimated statically for some number of operating conditions. But you need adaptive algorithms for beamforming, beamsteering, and digital predistortion so that the system can be continuously adjusted for the operating conditions at any given point in time.
That's addressed by a combination of static analysis for the characterization of the impairments and the development of models and simulation to understand the system's architecture. Only then can we assess how the system behaves in the presence of wideband signals.
Finally, a progression of techniques can be used to develop and test acquisition algorithms. The last step would involve prototyping using either software-defined radios or FPGAs that can be targeted with your algorithms. But one should always start with static analysis.