This paper presents a study of several implementations of the Mueller and Muller symbol rate timing recovery algorithm (1976) for ISDN transmission over digital subscriber loops (DSL). Implementations of this algorithm using various estimates of a specified timing function are investigated. It will be shown that despite the fact that all of the estimates considered are derived based on one set of conditions, their performance varies widely in a real system. The intrinsic properties of these estimates are first analyzed, then their performance on real subscriber loops is studied through extensive simulations of a practical digital receiver. The effect of various system parameters such as channel distortion and additive noise are included. Possible sources of convergence problems are also identified and corrective action proposed.>

Continuous phase modulation (CPM) schemes, using Nyquist 3 pulse shaping and correlative codes, are popular candidates for mobile radio applications. This paper presents a detailed study of the effects of Nyquist 3 pulse shaping on the spectral characteristics of CPM schemes with a modulation index of 1/2. The influence of pulse parameters such as its shape, bandwidth, coding polynomial, and truncation length on spectral characteristics such as main lobe, fractional power bandwidths, and out-of-band power levels is thoroughly investigated; error performance is also briefly analyzed. Among uncoded schemes, a new scheme using a pulse satisfying both Nyquist 3 and Nyquist 2 criteria is pointed out as a promising candidate and as a good compromise between spectral and power efficiencies.

It is shown that a band-limited Nyquist 3 pulse is expressible as a weighted infinite sum of delayed versions of a corresponding Nyquist 1 pulse. Weightings are derived explicitly for the minimum-bandwidth case. Also, a baseband system using Nyquist 3 shaping and integration at the receiver (over one symbol interval) is studied. An optimal band-limited Nyquist 3 system using integration at the receiver is shown to have the same error performance as that of an optimal Nyquist 1 system which uses sampling at the receiver.