Volterra series expansion is a widely used method for modeling nonlinearly distorted systems because of its simplicity, linearity with respect to the kernel coefficients, ability to generalize any order of nonlinearity, which makes it a suitable analytical tool for a wide range of applications. However, the higher computational complexity of the kernels is a major factor that limits its use in real-time applications. The number of coefficients increases exponentially with the sample memory length and the order of nonlinearity, increasing the resource utilization immensely and restricting the system bandwidth. In this paper, we have discussed different architectures such as a direct form (with modification for timing improvement) and an eigenvalue decomposition based parallel-cascade method for implementation of a low cost real-time quadratic Volterra filter. Both the designs are analyzed based on their resource utilization and timing constraints on a Xilinx Zynq-7000 SoC FPGA. Finally, the performance of the nonlinear filters for image reconstruction in real-time are demonstrated with a quality assurance ultrasound phantom data, sampled at a rate of 40 MHz.