Self-interference is the primary obstacle to full-duplex wireless communication on a single frequency band. This paper presents a novel convex reformulation for tuning the attenuation and phase shift parameters of a multiple-tap analog self-interference canceller. The standard optimization formulation for a multiple-tap analog canceller is non-convex, causing gradient descent algorithms to converge to local optima. By exploiting the architecture of an ideal analog canceller, an analytical solution for the global optimum is derived that minimizes mean-square error between the canceller and negative channel frequency responses. Multiple taps are necessary to emulate practical antenna coupling channels over a broad bandwidth. Simulated and measured results are presented that achieve significant wideband cancellation in multipath channel environments.