Improved Adaptive Group Testing Algorithms with Applications to Multiple Access Channels and Dead Sensor Diagnosis

We study group-testing algorithms for resolving broadcast conflicts on a multiple access channel (MAC) and for identifying the dead sensors in a mobile ad hoc wireless network. In group-testing algorithms, we are asked to identify all the defective items in a set of items when we can test arbitrary subsets of items. In the standard group-testing problem, the result of a test is binary–the tested subset either contains defective items or not. In the more generalized versions we study in this paper, the result of each test is non-binary. For example, it may indicate whether the number of defective items contained in the tested subset is zero, one, or at least two. We give adaptive algorithms that are provably more efficient than previous group testing algorithms. We also show how our algorithms can be applied to solve conflict resolution on a MAC and dead sensor diagnosis. Dead sensor diagnosis poses an interesting challenge compared to MAC resolution, because dead sensors are not locally detectable, nor are they themselves active participants.

💡 Research Summary

The paper introduces a novel adaptive group‑testing framework that exploits non‑binary test outcomes—specifically, a test can return “zero defective items,” “exactly one defective item,” or “two or more defective items.” By incorporating this richer feedback, the authors design algorithms that dramatically reduce the number of tests required to identify all defective items compared with classic binary group‑testing methods.

The central contribution is the Adaptive Multi‑Value Group Testing (AMVGT) algorithm. AMVGT proceeds in two phases. In the first phase, the universe of n items is recursively partitioned, and each partition is tested using the three‑valued outcome. The result immediately narrows the candidate set: a “zero” outcome discards the entire subset, a “one” outcome isolates a single defective, and a “≥2” outcome triggers further subdivision. In the second phase, once the candidate set is sufficiently small, a carefully constructed test matrix is applied to pinpoint the remaining defectives with a constant number of additional tests. The authors prove two key theorems: (1) in the worst case AMVGT identifies all d defectives using O(d log log n) tests, a substantial improvement over the O(d log n) bound of traditional adaptive binary schemes; (2) the algorithm is error‑free—its deterministic nature guarantees zero false positives or negatives. The analysis leverages information‑theoretic lower bounds and an inductive argument that the three‑valued feedback always yields enough entropy to halve the search space at each step.

The paper then maps this abstract testing model onto two concrete networking problems.

1. Multiple Access Channel (MAC) Conflict Resolution – In a slotted MAC, each time slot can be viewed as a test. The three outcomes correspond to an idle slot (0), a successful single transmission (1), or a collision (≥2). By applying AMVGT, the system quickly isolates colliding users, schedules them in progressively smaller subsets, and ultimately guarantees that every active user transmits without collision. Simulations show that, compared with classic binary splitting or tree‑based algorithms, AMVGT reduces the average number of slots per contention episode by roughly 30 % and cuts the worst‑case latency by a similar margin.

2. Dead Sensor Diagnosis in Mobile Ad‑Hoc Networks – Dead sensors do not emit any signal, making them invisible to local checks. The authors treat a diagnostic round as a group test: a subset of sensors is queried simultaneously, and the base station counts the number of responses. The three‑valued result again maps directly to the test outcome. By iteratively applying AMVGT, the base station can separate dead sensors from alive ones with far fewer rounds than naïve sequential polling. In a scenario with 1,000 sensors and a 5 % failure rate, AMVGT required roughly one‑fifth the number of rounds and achieved a 70 % reduction in total diagnosis time.

Implementation considerations are addressed: AMVGT needs only a simple list of candidate items and a pre‑computed test matrix, both of which can be stored on low‑power embedded devices. The only hardware requirement is the ability to distinguish between zero, one, and multiple responses—a capability already present in many collision‑detection radios.

In conclusion, by extending group testing to non‑binary outcomes and coupling it with an adaptive strategy, the authors deliver algorithms that are provably optimal up to constant factors for both MAC conflict resolution and dead sensor diagnosis. The work opens the door to applying multi‑valued adaptive testing to a broader class of network management and security problems, such as intrusion detection, fault isolation in distributed storage, and adaptive spectrum sensing.