Molecular communication via diffusion (MCvD) suffers from significant inter-symbol interference (ISI) due to the stochastic nature of molecule propagation. To address this challenge, this paper proposes a novel ternary modulation algorithm, JT-MoSK (Joint Threshold-based Molecular Shift Keying), which combines dynamic threshold detection with a sliding window mechanism to enhance demodulation reliability. The key innovation lies in a dual-stage detection process where each symbol undergoes two independent verifications: first as the current symbol and then as the preceding symbol in the next time slot. By introducing a ternary molecular encoding scheme (A/B/C) with strict alternation constraints, the algorithm leverages progressive error correction—initially selecting the second-most probable molecule type upon conflict and ultimately defaulting to the least probable type if necessary. A dynamic reliability threshold (τ), adaptively calculated from channel conditions, enables efficient classification of demodulation confidence into high, medium, or low levels. Simulation results confirm that the proposed JTMoSK scheme significantly outperforms conventional modulation methods in BER performance while maintaining computational efficiency. This performance advantage stems from the innovative combination of ternary molecular encoding and the dual-stage verification mechanism, which effectively balances detection accuracy and implementation complexity in diffusion-dominated environments.

In this study, a family of time‐invariant rate‐compatible (RC) low‐density parity‐check (LDPC) convolutional codes with algebraic construction is developed based on the graph‐extension method over a given positive integer m , which is considered as the given encoding memory. The base code of this family is first obtained by modifying a given construction of a time‐invariant LDPC convolutional code. Afterwards, the authors derive the other extended component codes by successively adding extra information bits into the base code. In particular, each component code in the family not only enables the property of fast encoding that can dramatically reduce the implementation complexity, but has the given encoding memory m for each parity‐check bit that can remarkably improve the error performance as well. They also demonstrate that the Tanner graph for all the codes in the family is free of girth 4. Simulation results indicate that the proposed family of RC‐LDPC convolutional codes collectively offers a desirable performance improvement as compared with the existing counterparts over the binary‐input additive Gaussian noise channels under BP decoding.