Capacity penalty due to ideal zero-forcing decision-feedback equalization

We compare the capacity CZF of an idealized zero-forcing decision-feedback equalization (ZF-DFE) system to the capacity C of the underlying Gaussian-noise channel. We find that, for strictly bandlimited channels, the capacity penalty gets vanishingly small as the signal power increases. For non-strictly bandlimited channels, on the other hand, the capacity difference C − CZF grows without bound as the signal power increases, and the asymptotic ratio CZF ⁄ C is as low as 93.6% for some channels. We conclude that ZF-DFE on non-strictly bandlimited channels is a suboptimal, information-lossy equalization technique, even in the limit of infinite signal power.