| We propose multi-sensor adaptive algorithms that separate cochannel signals in a mobile telephone network. Three signalling formats are considered: analog angle modulation (FM/PM), steady (nonbursty) digital signalling, and bursty digital signalling. All the methods are based on the beamform-and-cancel principle, whereby a cochannel signal is beamformed/isolated and is removed from the input mixture, thereby iteratively allowing the isolation of several sources. Each of the proposed architectures may be regarded as being composed of many cascaded stages, each of which is composed of a beamformer that isolates a single cochannel source, and a canceler that removes the isolated source from the input mixture.;For analog signals, we adopt the constant modulus (CM) array at the beamformer of each stage. We show that the mean-square error (MSE) of the beamformer in any stage is dominated by its adaptation step-size, and by that of the canceler in the preceding stages. We also prove that under certain conditions of SNR and angular separation, cascade processing has little negative effect on the ability of a stage to estimate the direction of a cochannel source.;We extend the multistage CM array to steady digital sources that possess the CM property. Since digital sources exhibit intersymbol interference (ISI), the corresponding separation architectures use additional adapting elements that exploit the temporal spread of the signal energy caused by ISI. We analyze one such architecture and show that at the first-stage output, the power of the captured source is reduced by a factor proportional to the inverse of the SNR. This near-complete cancelation of the first source at the output of the first stage enables the second stage to capture another cochannel source, and to do so with a performance almost as good as that of the first stage.;We use the cellular telephone standard IS-54 (or equivalently, its successor standard IS-136) to develop signal separation algorithms for bursty digital signals. Unlike the previous two signalling formats, IS-54 has a synchronization sequence that may be used to train a beamformer to reject cochannel interference. However, a given slot may face cochannel interference which does not intersect its training sequence. We exploit the beamform-and-cancel principle to derive an algorithm that overcomes this problem by block processing the data over two passes. The forward pass beamforms those regions of a slot where the cochannel interferers overlap the training sequence. The reverse pass then beamforms over the remaining regions. |
