We propose an algorithm for the joint estimation of frequency and frequency-rate in signals affected by fading and AWGN. Instead of processing the raw data directly, a stochastic nonlinear filter (NLF) is used for sequential phase estimation; the concatenated ML frequency and frequency-rate estimators are then fed by the NLF phase outputs. As a result of employing stochastic filtering the ML estimators exhibit enhanced robustness to changes of the signal-to-noise ratio and smaller threshold effects. An application of the proposed algorithm to the quasi-coherent detection of M-DPSK signals in nongeostationary satellite communications is also presented. Simulations show that the proposed receiver is rather insensitive to (time-varying) Doppler frequency shifts in flat-fading Rician channels.