Machine Learning for Automating the Design of Millimeter-Wave Baluns
Abstract : We propose a framework to analyze mm-wave baluns directly from physical parameters by adding a dimension of Machine Learning (ML) to existing electromagnetic (EM) methods.From a generalized physical model of mm-wave baluns, we train physical-electrical Machine Learning models that both accurately and quickly compute the electrical parameters of mm-wave baluns from physical parameters, reducing the need for full-wave simulations and advancing several aspects of mm-wave designs. One of the advancements is a fully automated design process that accurately generates full EM designs of mm-wave baluns when given an electrical specification and a metal option. The automated technique only takes several seconds to complete, compared to hours-weeks of the current trial-and-error methods, and notably the approach can optimize mm-wave baluns directly for the lowest metal loss. Another advancement is the theoretical interpretation of several high-level and abstract questions concerning mm-wave designs, in which we quantify the optimum transistor sizes for the last stage of a class-AB differential power amplifier on an on-chip process and derive the rule of thumb describing the inverse relationship between the optimum device sizes and mm-wave frequencies.
• We introduce a time-adaptive learning mechanism to cope with the non-stationarity of energy traces due to gain control adjustments and node movement.
• This run-time adaptation is barely explored in specialized work in the field of statistics but is critical for our approach.
• It also sets our scheme apart from existing work based on simple clustering or thresholding, which is highly sensitive to non-stationary behavior and thus often fails.
• We extend the learning framework to jointly process mm-wave channel traces from multiple time-synchronized sniffers.
• The idea is that different viewpoints of the same channel can provide diverse information and lead to higher decoding accuracies
The deployment of the next generation of wireless communication has drawn researchers and engineers together at this frequency band. While RF circuit design often utilizes inductors and transformers, mm-wave design increasingly employs Microwave circuits. Driven by emerging mm-wave challenges, various microwave topics have emerged, and numerous microwave solutions have been proposed to resolve mm-wave challenges. In this thesis, starting from the theory of coupled lines, we develop a common class of solutions for mm-wave EM designs that can resolve many emerging mm-wave challenges, which include Impedance Transforming Baluns, Power Combiners, Out-phasing circuits, and Doherty networks.
The proposed technique utilizes a combination of template matching and an Explicit Duration Hidden Markov Model (EDHMM) to correctly classify frames, while coping with the non-stationarity of the traces. This leads to a protocol level monitor that does not need to decode the channel at the physical layer, but just infers the type of packets that are exchanged based on sub-sampled energy traces. The performance of this framework is evaluated using off-the-shelf mm-wave wireless devices, quantifying its detection performance in the presence of one or multiple sniffers, and assessing the impact of physical layer parameters such as noise power and signal levels.
? Always the fundamental problem of any wireless communication system, generating more output power is often achieved by employing Power Combiner structures, since this approach can boost the output power without trading off linearity performance.
? At mmwave frequencies, generating power is even harder because the mm-wave range is close to the cut-off frequency of transistors.
? T-Line parallel combiners are the prevalent choice, but to design series-connected power combiners at mm-wave frequencies remains a major challenge.
|