This study addresses the limited adaptability of Bundle Protocol/Licklider
Transmission Protocol (BP/LTP) parameter configuration in deep-space
communication environments through a cooperative optimization mechanism that
operates under dynamic link conditions and diverse service demands. Conventional
BP/LTP parameter settings typically rely on fixed thresholds or empirical rules,
assuming relatively stable link states. However, in practical deep-space
missions, the bit error rate, propagation delay, and available transmission rate
all exhibit significant temporal variability. Moreover, variations in data size
and reliability requirements further exacerbate the performance degradation
caused by parameter mismatches. Owing to the multi-layer encapsulation and
resource coupling of bundle size, block size, segment size, and the number of
sessions, inappropriate parameter configurations can result in increased
acknowledgment and retransmission overhead, underutilized link capacity, and
prolonged end-to-end delivery latency. Traditional approaches are thus
inadequate for achieving holistic, real-time, and joint optimization in such
dynamic environments, necessitating adaptive parameter tuning mechanisms with
link-state awareness.
This paper formulates BP/LTP parameter optimization as a sequential decision
process and proposes a method based on a Deep Q-Network (DQN) for adaptive
parameter configuration. The state space includes transmitted data size, bit
error rate, one-way propagation delay, and available transmission rate. Based on
this state information, the DQN selects the bundle size and LTP block size from
predefined candidate sets. The LTP segment size is derived from a mapping rule
related to link conditions, and the number of concurrent sessions is calculated
online using an analytical model. The method is evaluated on a satellite–ground
simulation platform that consists of a BP/LTP sender, a BP/LTP receiver, and a
computing unit for DQN parameter optimization. It is compared with three fixed
parameter configurations in 36 experimental scenarios. The results show that,
compared with the fixed configurations, the proposed method reduces end-to-end
delivery latency, improves link utilization efficiency, and maintains more
stable transmission performance under different link conditions and data
scales.