Robotic computer–aided tuning of multi-cavity RF filters

摘要

Due to the demand for precision RF filter solutions, fully automated cavity filter manufacturing systems are a topic of interest for researchers. Currently, tuning stages for filter production lines are implemented by hand. This stringent process is both expensive and time consuming. Depending on the complexity of the cavity filter, this process may take up to several hours. Therefore, it is not suitable for higher volume production. To overcome this problem, Radio Frequency Systems (RFS Pty Ltd.) company is trying to develop a number of automated filter manufacturing systems that make the leap from conventional ‘trial and error’ manual filter tuning to automatic robotic tuning and set a new standard for filter production. The aim of this project, supported by RFS, is to design and manufacture a Robotic Computer-Aided Tuning (RoboCAT) system. The first section of this thesis deals with the design and fabrication of an automated robot arm to interface with tuning elements. For this purpose, a customised coaxial screw/nut driver is created to tune and lock the tuning elements simultaneously. Aspects of the automated tuner include: (1) Backlash compensation to increase the tuning resolution and reduce the tuning time. (2) Absolute positioning in order to have a second feedback source for robot codes along with obtained data from Computer–Aided Tuning (CAT) software. (3) Fault recognition ability to detect any potential error in CAT codes in early stage. The second stage of this project deals with finding a proper tuning instruction. By having a complete tuning instruction, it is feasible to write a standalone tuning code for an automated tuning system. It can be concluded from literature that circuit model parameter extraction is the only ideal tuning technique to be implemented by automated setup. This technique allows all elements to be tuned simultaneously rather than sequentially. However, in order to tune the filter using this technique, adequate initial optimization variable values are required to prevent the system from running into local minimum or failing to converge to the proper solution. This case arises when the filter is highly detuned. To overcome this problem, a coarse tuning technique based on phase format of input reflection coefficient of the filter is proposed in this thesis. In this method, resonators are tuned by bringing successive resonators to resonance, while the phase passes through the ±180˚ and 0˚ crossing at the center frequency. At the end of each sequence cross coupling is mapped across the entire range of its motion. Tuned cross coupling is recognized by measuring the return loss of the filter. Written codes based on this technique are able to guide the robot through coarse tuning process in a short time. Then, the circuit model parameter extraction technique is practically implemented by automated setup to complete the tuning. This proposed tuning system has been validated through experimental results. Results showed that utilizing the backlash compensation solution enables the robot to use accurate ‘1’ arcminute rotational resolution to achieve less than ‘5’ KHz frequency deviation in obtained filter response. This cannot be achieved with manual tuning. Elimination of mapping the tuning elements throughout the lash reduced the overall tuning time by 11 minutes and 23 seconds in an average of twenty tuning attempts. Absolute positioning system of the RoboCAT successfully detected the faulty tuning attempts caused by an error in CAT software. This enabled the robot to be function without supervision. Fabricated comprehensive coaxial/screw nut driver fitted with the designed SCARA, Cartesian, and multi-armed robots in order to tune the different filter types in the company without need for design updates. Several tuning attempts have been performed by the automated setup utilizing the proposed coarse tuning technique. Obtained filter response at the end of each tuning was very close to the ideal filter response. Therefore perfect initial values for the variables to be optimized were provided for the fine tuning program. This reduced the fine tuning time from 13 minute to less than 32 seconds and prevented the system from running into local minimum. Successful results in all tuning runs showed effectiveness of the tuning technique. Obtained tuning times via created setup was compared with traditional manual tuning attempts. RoboCAT achieved a tuned filter with an average time of 6 minutes compared to the manual tuning approach which took 42 minutes. These results obtained from 20 tuning attempt on a six-pole cross-coupled filter while the filter was tuned to channel 40 at a center frequency of 613.5 MHz.