This paper deals with the joint symbol-channel estimation for quasi-synchronous Multi-carrier Direct-Sequence Code DivisionMultiple Access (MC-DS-CDMA) systems over Rayleigh fading channels. To solve this non-linear problem, Rao- Blackwellized particle filters have proved efficient. In this framework, our contribution is twofold. 1) Instead of using an autoregressive (AR) model which does not match the bandlimitation of the theoretical power spectrum density (PSD) of a Rayleigh channel, we suggest modeling the channel by a low-pass filtered version of the so-called stochastic sinusoidal process. It consists of sinusoids in quadrature with random magnitudes modeled as AR processes. By suitably choosing the AR parameters, this combination has the advantage of providing a stationnary process whose PSD is bandlimited and has two peaks at the maximum Doppler frequency for any AR order. 2) The estimation of the model parameters is included in the joint symbol-channel estimation process. For this purpose, a deterministic Rao-Blackewellized Particle Filter which jointly estimates the symbols and the channels is cross-coupled with a Kalman Filter which yields the AR parameters.