Executable Code Information Hiding via Reversible Instruction Swapping

The patent changes the parameters of instruction embedding by using multiple interchangeable instruction pairs with different operational characteristics (arithmetic, logic, data movement). Instead of relying on simple semantic equivalence, the method varies instruction parameters such as opcode types, operand structures, and execution semantics while maintaining functional equivalence. This makes detection more difficult because the embedded data is distributed across diverse instruction types rather than a single pattern.

The patent introduces a new dimension to instruction swapping by organizing interchangeable instructions into matrix structures where rows represent different instruction pairs and columns represent different embedding positions. This matrix-based approach adds structural complexity beyond simple pairwise substitution, creating a multi-dimensional embedding space that is harder to analyze and detect through traditional static analysis methods.

The patent performs preliminary analysis to identify all valid interchangeable instruction pairs before the embedding process. By pre-computing and storing the set of valid swap locations and their corresponding instruction pairs, the method eliminates the need for complex real-time analysis during data embedding. This preliminary action creates a lookup table of valid positions, simplifying the actual embedding operation while maintaining comprehensive coverage of possible swap locations.

The patent introduces an intermediary matrix structure that mediates between the hidden data and the instruction swaps. The matrix serves as an intermediate representation that maps data bits to specific instruction pair selections and swap positions. This intermediary layer simplifies the embedding process by providing a systematic method to translate arbitrary data into a sequence of valid instruction swaps without requiring complex optimization or analysis at embedding time.

The patent changes the parameters of the embedding process by using random matrix generation with specific mathematical properties (full column rank). Instead of using fixed or deterministic patterns to select instruction pairs, the method employs randomly generated matrices where the statistical properties provide unpredictability. The full column rank requirement ensures mathematical validity for data recovery while the randomness enhances undetectability. This parameter change from deterministic to probabilistic selection is key to improving security.

The patent creates multiple copies of the embedding structure through generating multiple random matrices, each capable of independently embedding and recovering data. Rather than relying on a single complex embedding scheme, the method produces a set of equivalent matrices that can serve as backups or alternatives. This copying approach distributes the complexity across multiple simpler structures, making the overall system more robust and harder to analyze while maintaining the same functional capabilities.