TASPER: Efficient TWT Scheduling Resolver

• TASPER is a heuristic resolver for the NP-hard TWT Acceptance and Scheduling Problem, using a directed decision graph to schedule non-overlapping TWT service periods in Wi‑Fi networks.
• It optimizes transmission admission and sequence selection by balancing rejection penalties and energy costs through a tunable parameter, β, while respecting AoI deadlines.
• Empirical results show TASPER achieves near-optimal rejection cost and significant energy savings compared to baselines, validated in both simulations and IIoT testbed experiments.

Searching arXiv for the TASPER paper and nearby acronym collisions. TASP Efficient Resolver, abbreviated TASPER, is the heuristic strategy introduced for the TWT Acceptance and Scheduling Problem (TASP) in time-sensitive, energy-efficient Wi‑Fi networks. In the formulation of the underlying problem, an access point schedules non-overlapping TWT Service Periods (SPs) within each beacon interval so as to maximize traffic throughput and energy efficiency while respecting Age of Information (AoI) constraints. Because TASP is NP-hard, TASPER is proposed as a near-optimal resolver that searches a restricted decision graph of feasible transmission orders, balancing transmission admission and station energy cost through a tunable trade-off parameter $\beta$ (Busacca et al., 30 Sep 2025).

1. Terminology and scope

In the TASPER paper, TASP denotes the TWT Acceptance and Scheduling Problem, and TASPER denotes the TASP Efficient Resolver introduced to solve it heuristically (Busacca et al., 30 Sep 2025). The setting is IEEE 802.11 Target Wake Time (TWT) for Industrial Internet of Things (IIoT) deployments, where an access point assigns wake times and service durations to stations so that they transmit in contention-free, non-overlapping intervals.

The core motivation is that Wi‑Fi has historically been considered unsuitable for Time Sensitive Networking because contention and collisions lead to nondeterministic delay. The paper argues that this limitation can be mitigated by using TWT so that stations wake only during assigned service periods and sleep otherwise, thereby enabling centralized, slot-based scheduling with explicit AoI control (Busacca et al., 30 Sep 2025).

The acronym TASP is used differently in other arXiv literature. In "TASP: Topology-aware Sequence Parallelism" (Wang et al., 30 Sep 2025), TASP refers to a communication method for long-context LLM attention, and the paper explicitly does not define or mention “TASPER.” In "Tightest Admissible Shortest Path" (Weiss et al., 2023), TASP refers to a shortest-path problem under bounded uncertainty, and that paper likewise does not use the term “TASPER.” This suggests that, in current arXiv usage, TASPER is specifically grounded in the Wi‑Fi TWT scheduling context of (Busacca et al., 30 Sep 2025).

2. Problem formulation: the TWT Acceptance and Scheduling Problem

The TASP formulation assumes one access point and $M$ stations. Let $N$ denote the set of transmission requests collected at the beginning of a beacon interval. Each transmission $j$ is characterized by a station identifier, payload size $b_j$, generation time $g_j$, AoI deadline $d_j$, and priority $p_j$. If station $m$ experiences link quality $l_m$, then its transmission duration is modeled as

$M$0

All of $M$1, $M$2, and $M$3 are expressed in slots (Busacca et al., 30 Sep 2025).

The decision variables consist of an admission vector $M$4, a sequence matrix $M$5, and an end-time vector $M$6. The model also includes dummy start and end transmissions, $M$7 and $M$8, with null duration and deadline equal to the maximum deadline among real transmissions. TASP then jointly determines which requests are accepted and in what order they are scheduled (Busacca et al., 30 Sep 2025).

The objective combines rejection cost and energy cost through a weighting parameter $M$9. In the notation of the paper, the optimization minimizes a weighted sum in which rejected transmissions incur penalty $N$0, while admitted transmissions incur an energy term dependent on their predecessor in the schedule (Busacca et al., 30 Sep 2025). The formulation is subject to sequence feasibility, non-overlap, release-time, and deadline constraints, including:

$N$1

and

$N$2

The energy model distinguishes transmission, idle, and mode-transition energy. If $N$3 indicates that transmissions $N$4 and $N$5 belong to the same station, then the cost of scheduling $N$6 immediately after $N$7 is given by

$N$8

This expresses the fact that, for consecutive transmissions of the same station, the device may either remain idle or return to sleep and wake again, whichever is cheaper (Busacca et al., 30 Sep 2025).

The paper states that TASP can be mapped to a single-machine, sequence-independent, completion-dependent batch setup cost scheduling problem with release dates, deadlines, and rejection, and explicitly proves that TASP is NP-hard (Busacca et al., 30 Sep 2025).

3. TASPER algorithmic design

TASPER is described as a heuristic inspired by the BALAS algorithm for Order Acceptance and Scheduling. Its central abstraction is a directed decision graph in which a path from dummy start $N$9 to dummy end $j$0 corresponds to a feasible transmission schedule (Busacca et al., 30 Sep 2025).

Vertices consist of $j$1, one vertex per transmission whose AoI deadline has not expired, and $j$2. Edges connect feasible predecessor-successor pairs. Each edge $j$3 carries a bi-dimensional weight

$j$4

where $j$5 is the duration of transmission $j$6, and the edge value is

$j$7

This ranking favors high-priority transmissions with low energy cost (Busacca et al., 30 Sep 2025).

A candidate transmission $j$8 is feasible after a partial path ending at time $j$9 only if

$b_j$0

Thus, the algorithm respects both release times and AoI deadlines during path extension (Busacca et al., 30 Sep 2025).

To reduce search complexity, TASPER restricts exploration to a neighborhood. All unscheduled, non-expired transmissions are first sorted by latest start time $b_j$1. Transmission $b_j$2 lies in the neighborhood of transmission $b_j$3 if their indices in that ordering differ by at most $b_j$4, where $b_j$5 is a tunable neighborhood size. The paper uses $b_j$6 in simulation as a practical trade-off between search depth and runtime (Busacca et al., 30 Sep 2025).

At each step, TASPER chooses a dominant next transmission from the neighborhood: first by maximum edge value, then by shortest completion time under ties. Partial paths are recursively extended and pruned using a dominance test among paths that end at the same transmission. The best path is then the one with maximum cumulative reward (Busacca et al., 30 Sep 2025).

The algorithm’s complexity is given as

$b_j$7

where $b_j$8 is the number of transmissions and $b_j$9 is the maximum number of transmissions simultaneously ready for scheduling. For fixed $g_j$0, if each of $g_j$1 stations generates $g_j$2 packets per beacon interval, complexity becomes

$g_j$3

which the paper presents as linear in the number of stations (Busacca et al., 30 Sep 2025).

4. Relation to baseline schedulers

The principal baseline is ShortestFirst, which repeatedly selects the feasible transmission with smallest duration $g_j$4, breaking ties by earliest AoI deadline, then highest priority, then oldest generation time. The paper identifies this as the strongest baseline in most comparisons because short transmissions tend to preserve airtime and save energy. TASPER outperforms it by using path-based lookahead rather than purely local duration minimization (Busacca et al., 30 Sep 2025).

Other baselines are FIFO, PriorityFirst, and Random. FIFO orders transmissions by generation time and drops expired requests. PriorityFirst prefers highest-priority traffic first, using deadline and duration as tie-breakers. Random selects among feasible transmissions uniformly at random. These baselines do not jointly optimize admission, ordering, and energy in the manner of TASPER (Busacca et al., 30 Sep 2025).

The paper also adapts DC-HSA, a WirelessHART scheduling method, to the Wi‑Fi uplink setting. In the adapted version, the network topology becomes single-hop star, the entire channel is allocated to the scheduled station, and the time interval is set to $g_j$5 ms. TASPER differs from HSA in that it explicitly optimizes a priority-energy trade-off, whereas HSA is fundamentally deadline- and priority-oriented (Busacca et al., 30 Sep 2025).

A plausible implication is that TASPER’s improvement derives not from a new PHY or MAC primitive, but from a better approximation to the mixed admission-and-scheduling objective under TWT’s contention-free execution model.

5. Empirical results

The main simulation study uses an extended ns-3-twt framework with support for AoI deadlines, multiple station energy models, and richer TWT logging. Scenarios include $g_j$6, $g_j$7, and $g_j$8 stations, one access point, 2.4 GHz, SISO 20 MHz, OFDMA enabled, and a beacon interval

$g_j$9

The evaluation spans 100 generated TASP instances over 100 consecutive beacon intervals (Busacca et al., 30 Sep 2025).

Against the exact TASP optimum computed by Gurobi for the 16-station case, TASPER is reported to be very close: with $d_j$0, mean rejection cost is only 0.04% higher than optimum, and with $d_j$1, energy is only 1.4% above optimum (Busacca et al., 30 Sep 2025).

In the larger 64-station scenario with $d_j$2, TASPER reduces mean rejection cost by 25% relative to ShortestFirst, 26% relative to HSA, 28% relative to PriorityFirst, 44% relative to FIFO, and 63% relative to Random. The paper’s abstract summarizes this as up to 24.97% lower mean transmission rejection cost than the leading baseline, ShortestFirst (Busacca et al., 30 Sep 2025).

Energy results show the expected $d_j$3-trade-off. For 16 stations, TASPER’s energy rises from 14.3 mJ at $d_j$4 to 17.4 mJ at $d_j$5, while for 64 stations it rises from 6.4 mJ to 11.2 mJ. Even so, the paper reports up to 14.86% more energy savings than ShortestFirst, and 34% lower energy consumption than HSA in challenging scenarios (Busacca et al., 30 Sep 2025).

In the highlighted 16-station simulation, deadline miss rates are 0.00% for TASPER, ShortestFirst, and HSA, while FIFO, PriorityFirst, and Random exhibit nonzero miss rates. The paper also states that packet loss rate is zero in all simulations because only one station is active per slot and durations are computed to fit capacity (Busacca et al., 30 Sep 2025).

6. Testbed validation, practical significance, and limitations

The paper validates TWT and TASPER on an IIoT testbed with 10 commercial TWT-compatible stations and one Wi‑Fi 6 access point. The stations are Espressif ESP32-C6-DevKitC-1 boards, and the access point is a Synology WRX560. TWT behavior is implemented using ESP-IDF, clocks are synchronized via Wi‑Fi TSF, and traffic is analyzed with Wireshark after capture by a monitoring host (Busacca et al., 30 Sep 2025).

A separate TWT-versus-non-TWT experiment shows that deterministic TWT scheduling reduces average median AoI from 58.2 ms to 49.1 ms, a 16% improvement, while reducing AoI standard deviation by 80%. Measured per-station power also falls from about 0.47 W to 0.24 W, corresponding to about 49% lower measured per-STA power (Busacca et al., 30 Sep 2025). This establishes the underlying contention-free TWT substrate on which TASPER operates.

In the direct TASPER-versus-ShortestFirst hardware comparison, TASPER is evaluated with $d_j$6 on a mixed traffic pattern containing both long and short flows. TASPER admits all STAs, with average AoI 22.05 ms, whereas ShortestFirst achieves lower average AoI, 11.66 ms, but fails to schedule a long, tight-deadline transmission from one station before its deadline (Busacca et al., 30 Sep 2025). This exposes the central design trade-off: lower average AoI for a subset of traffic versus broader admission of important transmissions.

The paper’s concluding summary states that TASPER admits more transmissions than the best baseline without violating any AoI deadline, whereas the detailed hardware subsection reports that one station exceeds deadline 5.8% of the time in that specific testbed scenario (Busacca et al., 30 Sep 2025). This suggests that the strongest no-deadline-violation claim is most clearly supported by the simulation results, while hardware behavior is affected by practical timing imperfections such as beacon overhead and TWT start imprecision.

Several assumptions delimit TASPER’s scope. The formulation focuses on individual TWT sessions, one station per service period, known link rates $d_j$7, negligible propagation delay, and no interfering networks in the main simulations (Busacca et al., 30 Sep 2025). The paper explicitly leaves OFDMA resource allocation across time-frequency, overlapping-BSS interference, and richer stochastic channel effects to future work. A plausible implication is that TASPER is best understood as a centralized, single-cell uplink scheduler for periodic or semi-periodic IIoT traffic rather than a complete general-purpose Wi‑Fi TSN stack.