The continual demand for increased mobile radio capacity prompts continued development of new technology to increase system performance. This thesis presents an investigation of the performance of a generalised linear diversity receiver, as an implementable approximation to the optimum diversity receiver. The receiver utilises space, or antenna, diversity combining to combat the effects of multipath fading experienced in the mobile channel. The receiver is intended for coherent reception of linearly modulated wideband time division multiple access (TDMA) digital transmissions. The receiver structure is also restricted to the linear case. Two major analyses are undertaken in this thesis: the performance of the diversity receiver with finite-length fractionally-spaced branch equalisers, assuming ideal channel-state-information and synchronisation; and the performance of a carrier frequency offset recovery algorithm, assuming only knowledge of the channel autocorrelation function, and burst synchronisation. This thesis presents a general time-varying frequency-selective multipath Rayleigh fading model. The system model includes non-white, non-Gaussian co-channel interferers as well as non-white, but Gaussian co-channel interferers. However, adjacent channel interference is not modelled. Simulation results are presented for the best possible performance (with ideal channel-state-information) of the diversity receiver, showing the effects of diversity and finite equaliser length, and are compared to results for the optimum linear receiver. The results show that increased diversity is very powerful in improving the performance of the receiver, and that the equaliser lengths can be quite short (e. g., 10 taps) and the receiver will still perform close to the optimum. The receiver is shown to be more capable of combating non-Gaussian co-channel interference than Gaussian (or noise-like) interference. The performance of the carrier frequency offset estimator is also determined by simulation. It is found that in most cases, doubling the diversity reduces the error standard deviation by a factor of two or more. The estimator is found to be relatively insensitive to symbol timing, channel delay spread, frequency offset and signal-to-noise ratio, especially with high diversity.