When spacecraft dynamics are described by a Linear Fractional Representation (LFR)—as is standard in robust AOCS design—analysing the estimation and control loops in isolation no longer reflects true closed-loop behaviour, because parametric uncertainties cause attitude and orbit determination errors to become intrinsically coupled with the control law. This paper addresses that gap by introducing a unified family of sensitivity functions that captures the coupled dynamics of attitude estimation and control actuation. Building on these joint sensitivity functions, a systematic framework for in-orbit system identification of the LFR has been developed. Multisine excitation signals are tuned by optimising the spectral properties of the Fisher Information Matrix (FIM), constructed directly from sensitivity gradients derived from the system LFR. An optimal experiment design is obtained by exciting the five structurally distinct injection points available in the AOCS architecture. Together, these results enable rigorous, frequency-domain verification of stability margins, knowledge error budgets, and pointing error budgets directly from flight data.