ORIGINAL ARTICLE Design of custom-made stacked patch antennas: a machine learning approach Satish K. Jain • Amalendu Patnaik • Sachendra N. Sinha Received: 27 July 2011 / Accepted: 16 February 2012 / Published online: 6 March 2012 Ó Springer-Verlag 2012 Abstract Machine learning approaches, viz. particle swarm optimization in conjunction with neural networks have been used to develop a user friendly tool to design custom-made stacked patch antennas in the entire X-Ku band for satellite communication application. The role of the neural network is to develop a black-box model to relate the frequencies of operation of the antenna with its dimensional parameters. This trained neural network was embedded in the loop of a particle swarm optimizer to decide the design dimensions required for specific user defined frequencies. The effectiveness and accuracy of the developed model are verified by simulations and experi- mental measurements. Keywords Neural network  Particle swarm optimization  Stacked patch antenna 1 Introduction In response to the ever-increasing needs of antenna band- width, considerable amount of effort is currently underway to design multiband antennas. Mostly planar designs are preferred for these structures for their added advantage of small size, low manufacturing cost and conformability. In microstrip antenna technology, stacked configuration is one of the easy and viable solution for getting multiband characteristics, because in this, the size of the structure doesn’t increase in planar direction, giving flexibility in making microstrip antenna arrays. But with the increase in the number of layers in stacked configuration, the design parameters deciding the frequency of operation of these antennas, also increases. From designers’ point of view, it becomes a difficult task to decide the optimized values of all the design parameters for the antenna to have specified operating frequencies. Commercially available packages for microstrip antenna design are computer intensive and require large computer resources. To reach to a final optimized structure, it might need several simulations. In order to reduce this time of computation, some commer- cially available packages are now available with optimiz- ers, but for this also, number of simulations are required. Stacked patch antennas have been investigated exten- sively [1–7]. Depending on the size of the patches and other design parameters, these antennas can be used either as a multiband antenna, when it resonates at two different frequencies; or as a broadband antenna when the resonating frequencies are very much close to each other [7]. In other words, the type of resonance in a stacked patch antenna depends on all its design parameters. Although some design procedures are available in the literature [4–6], there is still need for a more user-friendly approach from application and design point of view. In the present study, instead of analytical approaches, machine learning approaches have been used for design of stacked patch antennas in X-Ku band for satellite appli- cations. Particle swarm optimization (PSO) in conjunction with neural networks (NN) technique is used to decide the design dimensions of the antenna, according to the requirements provided by the user. The role of the NN is to develop a black-box model between the design dimension S. K. Jain  A. Patnaik (&)  S. N. Sinha Department of Electronics and Computer Engineering, Indian Institute of Technology Roorkee, Roorkee 247 667, India e-mail: apatnaik@ieee.org S. K. Jain e-mail: satishjain.jain@ieee.org S. N. Sinha e-mail: sn_sinha@ieee.org 123 Int. J. Mach. Learn. & Cyber. (2013) 4:189–194 DOI 10.1007/s13042-012-0084-x