IC Substrate - Design automation for RF front-end optimization

When implementing carrier aggregation in the RF front-end, some new problems will be encountered, which are mainly related to the deployment of filters in an environment with dynamic impedance changes. This article will elaborate on some of these problems, and propose a design method to automatically process a large number of circuit-level candidate solutions, thereby reducing the burden on designers, and relatively easy to screen out feasible solutions.
Where are the design challenges?
Figure 1 is a simplified block diagram of a possible architecture of a bidirectional inter-band downlink carrier aggregation (DL CA) system, in which the RX branch of Band 3 can be dynamically connected in parallel with the TX/RX branch of Band 1. This design can be easily extended to multiple component carriers and different switching configurations. For example, "Infineon Mobile Communication Application Guide"recommends the use of single-antenna and dual-antenna downlink RFFE architectures, supporting up to 5 CA component carriers. The basic building blocks are switches, duplexers, and band-pass filters. These modules are well-known, high-quality, and widely used in mobile phones. Therefore, the current problem is in addition to intermodulation architecture selection and frequency planning In addition to the treatment, are there any issues that require special attention?
Unfortunately, the answer is yes: One of the major design bottlenecks is that once the filters are connected in parallel, they will inevitably affect each other's performance significantly. For example, Figure 2 shows that the band 8 and band 1 filters are connected separately and connected to Frequency response at public node. It is worth noting that the performance of the band 1 filter is completely destroyed by the band 8 filter, and the performance of the band 8 filter is basically unchanged due to the existence of the band 1 filter.
The out-of-band suppression performance of these two filters is very good, so even if there is filter leakage current, it cannot explain the damage of the band 1 filter. However, if we look at the input impedance of the Band 8 filter at the Band 1 frequency in Figure 3a, we can notice that the Band 8 filter looks like an open-ended transmission line with an electrical length of about 67 degrees instead of an open circuit. When the Band 1 filter is used to connect it to a common node, the Band 8 filter will load the performance of the Band 1 filter in a manner similar to an open stub, and this will completely change the performance of the filter!

At this time, we can already guess why the existence of the band 1 filter does not destroy the performance of the band 8 filter. If we look at the input impedance of the Band 1 filter at the Band 8 frequency (Figure 3b), we will find that the Band 1 filter is essentially an open circuit, which is purely coincidental. Knowing this, we can conceive a practical goal, which is to design a matching circuit (phase shifter) to preserve the passband behavior of the filter while mapping the response of other component carrier frequencies to the open circuit. If this goal is successfully achieved, then the filters are equivalent to being transparent to each other and can be connected in any CA configuration. We call this part of the design process "filter matching"
The challenge of solving the filter matching problem
A matched solution can only become a more or less perfect solution under more occasional circumstances. This is usually the case for component carriers with wide frequency spacing, such as between a low-band (LB) pair and a high-band (HB) pair. When more than one component carrier frequency must be mapped to an open circuit, it is more difficult to achieve mutual open circuit. In addition, under the premise of not significantly affecting the passband behavior, component carriers of adjacent frequencies may be difficult to match. Another point is that there are usually conflicting constraints in practice, resulting in only a very small number of external matching components. Therefore, the ideal situation is to design the acoustic filter in advance so that it can be qualified for some CA schemes with very few matching components, but the filter itself still does not have enough design freedom to completely eliminate the need for external matching.
Therefore, our design process can still only try to match first. If it succeeds, we know that CA can basically work. In the design process of using filter cooperative matching, we often have to accept that the solution cannot provide a precise open circuit at the component carrier frequency, which leaves a lot of interaction and load between the filters. Referring to Figure 1, we also have switches that connect these interactions, and the electrical size of the switches is large enough, so they can also help effectively load one filter to another.
In short, to solve these problems together, it is necessary to fine-tune the complete model including switches, filters, and external matching circuits.
Example: Band 1 + Band 3 downlink carrier aggregation
The component carrier frequency bands are relatively close to each other. Use representative public domain S-parameter models for Band 1 duplexers and Band 3 RX filters, as well as general-purpose semiconductor SP2T models that support parallel throwing states. In the non-CA configuration, the switch connects the antenna to the Band 1 branch; in the CA configuration, the switch connects the antenna to the Band 1 and Band 3 branches. Therefore, the matching circuit should be optimized to make it suitable for these two configurations. We assign the switch RF1 node to band 1 and the RF2 node to band 3, and use the 0201 package size Murata discrete component model of the library LQW03AW_00 (inductor) and GJM03 (capacitor) to design the matching circuit.

We first try to match the band 3 filter. In all matching tasks, we use the RF design automation software platform OptenniLab, because it can automatically synthesize and optimize a large number of candidate topologies. This software is essential to our design: even if there are only 2 matching components at most, each circuit will have 17 different topology options, and when there is no obvious solution to achieve a good match, it is often difficult to predict which This combination of topologies can achieve the best performance. For example, for a single duplexer, if each branch has at most 2 matching components, there can be a total of 173=4913 different topologies. Most topologies are doomed to fail, but the RF design automation software platform can easily optimize and automatically sort more than 100 related topologies, while also taking into account the sensitivity of the solution to component tolerances. This greatly helped the design process, so that we would basically not miss the topological combination with the best performance and the strongest tolerance stability, otherwise it would be easy to miss such a solution if we only rely on manual derivation of a limited number of topologies.

Therefore, we use the band 3 filter model as the basis, and synthesize the matching circuit with the open circuit target of band 1 and the good insertion loss of RX in band 3 as the target. Since Band 1 and Band 3 are very close to each other, the common matching challenge we face is as follows: The frequency of Band 1 crosses a long arc on the edge of the Smith chart, and the results of trying to place it near the open circuit point are bound to match. The frequency band response creates a considerable compromise. There are many topological schemes to choose from, some of which have better insertion loss, and some can be better mapped to open circuits. It is difficult to have both. Figure 4 shows the impedances of Band 3 RX and Band 1, and compares the unmatched filter and our selected cooperative matched filter, including 3 matched components at the input of the filter and 2 components at the output.
This article compares two methods of matching CA filters. In the "cooperative matching" method, the filters are first matched individually, with the goal of achieving an open circuit at the frequency of another filter. After the results of these sub-problems are combined and fine-tuned, a feasible solution is usually obtained. However, this process basically can only get a matching topology, or it takes time and effort to manually combine the candidate results of each sub-problem. Therefore, we propose a second method called "full-image optimization", which omits the collaborative matching step and directly searches for the best circuit according to actual performance indicators (ie, signal insertion loss and suppression). In this way, the most economical solution can be identified very effectively. For the more complex CA architecture in practice, it may be more helpful to mix the two methods. We can use the "full graph optimization" design for some functional blocks, and then combine them and fine-tune them, similar to "collaboration" Match” method. In all these methods, the RF design automation platform we adopted plays a central role, because it eliminates most of the manual operations that designers must spend on designing software when solving CA problems.