Performance of the joint reduced rank model-based demodulator for asynchronous co-channel GMSK signals

Cellular communications systems suffer from co-channel interference (CCI), due to signals from adjacent cells interfering with each other. Conventional demodulation techniques have relied on interference rejection to extract only the highest power signal. Recently, a joint reduced rank model-based demodulator (J-RRMBD) was presented that extracts two or more co-channel signals using continuous phase modulation (CPM) simultaneously. The technique is based upon a root selection algorithm using forward-backward linear prediction (FBLP), whereby multiple unequal powered signals are extracted by choosing the highest powered roots. Using synchronous continuous phase frequency shift keying (CPFSK) signals, the algorithm has been shown to provide up to 5 dB improvement over the conventional quadrature demodulator (QD) and single user model-based demodulator (MBD), which extracts only the highest powered user. In this paper, we apply the technique to received asynchronous Gaussian Minimum Shift Keying (GMSK) signals with unknown frequency offsets, encountered in terrestrial GSM networks. We compare performance to a joint Viterbi demodulator when two co-channel signals are present and show that the new technique provides an order of magnitude reduction in bit error rate (BER), and can more reliably demodulate signals with few samples as long as the carrier-to-interference ratio (CIR) ≥ 3 dB. When CIR ≪ 5 dB, a single user version of the algorithm provides performance improvement up to 7 dB over the Viterbi algorithm.