A CLASSIFICATION–STRUCTURAL APPROACH TO RANDOMNESS EVALUATION OF SHORT BINARY SEQUENCES IN CRYPTOGRAPHIC PROTECTION SYSTEMS AND IOT TELEMETRY

Abstract

The article addresses the problem of evaluating the randomness of short binary sequences used in cryptographic protec- tion systems and IoT telemetry as keys, authentication tokens, service markers, and identifiers. It is shown that traditional sta- tistical randomness test suites (NIST STS, DIEHARD, TestU01) are designed for long samples and lose reliability when applied to sequences of 8–128 bits, which are typical for lightweight protocols such as ZigBee, LoRaWAN, RFID, and embedded con- trollers. To overcome these limitations, a classification-driven structural approach to randomness assessment is proposed, which relies on the analysis of non-overlapping k-bit blocks, the construction of empirical frequency vectors, and the use of a maximum deviation statistic from the theoretical distribution. A theoretical estimate of the probability of large deviations is derived using Hoeffding’s inequality, allowing one to specify a threshold value of the criterion for a predefined significance level and formally control the error probability in decision-making. The proposed approach formalizes the problem of randomness evaluation as a classification task, which makes it possible to compare generators and filter out non-random sequences based on quality metrics. The experimental framework includes four classes of sequence sources (algorithmic PRNGs, the cryptographic AES-CTR generator, and hardware noise sensors). For each source, short fragments of 8–128 bits were generated and evaluated in compari- son with ENT and basic NIST SP 800-22 tests using metrics such as classification accuracy, Type II error, ROC characteristics, and computational performance. The results demonstrate that the proposed method provides higher classification accuracy, superior ROC indicators, and lower variability of results on short fragments, while maintaining acceptable computational complexity for deployment on microcontrollers. Practical application scenarios are outlined for IoT sensors, smart home systems, embedded controllers, and telemetry-based IDS/IPS solutions, where the proposed criterion can serve as a lightweight randomness quality module and enhance the overall resilience of cryptographic protocols against the exploitation of structural defects in generators.