Low Power Wide Area (LPWA) networks are known to be highly vulnerable to external in-band interference in terms of packet collisions which may substantially degrade the system performance. In order to enhance the performance in such cases, the telegram splitting (TS) method has been proposed recently. This approach exploits the typical burstiness of the interference via forward error correction (FEC) and offers a substantial performance improvement compared to other methods for packet transmissions in LPWA networks. While it has been already demonstrated that the TS method benefits from knowledge on the current interference state at the receiver side, corresponding practical receiver algorithms of high performance are still missing. The modeling of the bursty interference via Markov chains leads to the optimal detector in terms of a-posteriori symbol error probability. However, this solution requires a high computational complexity, assumes an a-priori knowledge on the interference characteristics and lacks flexibility. We propose a further developed scheme with increased flexibility and introduce an approach to reduce its complexity while maintaining a close-to-optimum performance. In particular, the proposed low complexity solution substantially outperforms existing practical methods in terms of packet error rate and therefore is highly beneficial for practical LPWA network scenarios.