Det-5G: Closing the Determinism Gap in 5G-Advanced for Industrial Closed-Loop Control Adnan Aijaz
arXiv:2609.07386v1 [cs.NI] 7 Sep 2026
Bristol Research and Innovation Laboratory, Toshiba Europe Ltd., Bristol, United Kingdom [email protected]
while industrial guidance identifies non-deterministic timing as a production risk [5]. The issue is timely in 5G-Advanced: private 5G has enabled industrial connectivity, but adoption for production-critical control remains hindered by unpredictable command/feedback completion. This paper presents Deterministic-5G (Det-5G), a radio resource allocation solution that revisits industrial 5G scheduling with completion of the closed-loop control cycle as the primary objective. Its core design principle is that a command and its corresponding feedback form a single causally coupled transaction whose timing and reliability should be considered jointly, rather than as independent downlink and uplink deliveries. The paper makes the following contributions: • Cycle-level resource allocation and reliability control: a self-contained cyclic transmission procedure jointly provisions resources for command delivery, feedback, proactive redundancy, acknowledgement, and recovery. It supports both periodically reserved and cycle-by-cycle operation to balance signaling overhead and adaptability. • Single- and multi-device industrial control: the cycle-level design supports both individual control loops, representative of conventional industrial machines and production equipment, and coordinated groups of devices required by emerging applications such as wireless robot control, collaborative robotics, and distributed production systems. For multi-device operation, the framework incorporates common downlink delivery, selective recovery, and optiI. I NTRODUCTION mized sharing of uplink time-frequency resources. 5G has been positioned as a key wireless technology • Determinism-oriented performance evaluation: analytical for industrial automation because uRLLC, private/non-public and Monte Carlo methods evaluate complete-cycle behavnetworks, and time-sensitive communication extend cellular ior against dynamic grant-based scheduling, SPS/CG, and connectivity toward applications traditionally served by deterfixed proactive repetition, considering reliability, recoveryministic wired systems [1], [2]. Yet low packet latency and induced timing variation, multi-device scalability, airhigh reliability are not sufficient to make a wireless system interface granularity, and mobility. deterministic for control. A controller repeatedly issues a Det-5G uses existing New Radio (NR) mechanisms rather command and expects sensing or actuation feedback within than a new 5G-Advanced primitive. As 5G-Advanced evolves the same control cycle. The relevant performance object NR toward 6G [6], addressing the determinism gap within is therefore completion of a causally linked closed-loop the existing air interface becomes increasingly important for transaction, not latency of an isolated packet. Independent robotics and Physical AI [7]. downlink/uplink scheduling and reactive retransmissions can make cycle completion unpredictable even when average packet II. R ELATED W ORK latency is low. This gap matters for industrial adoption. Stringent motionEarly work on industrial closed-loop control focused largely control profiles combine millisecond-level transfer intervals and on non-cellular wireless systems. ENCLOSE introduced an latency with essentially no tolerance for missed cycles [3], [4], enhanced single-hop interface for factory automation, while
Abstract—The ultra-reliable low-latency communication (uRLLC) capability of 5G has created significant opportunities for industrial wireless connectivity, yet widespread use of cellular networks for closed-loop control remains challenging. Closed-loop control requires more than low packet latency and high reliability: cyclic command/feedback exchanges must complete within predictable time bounds despite changing channel conditions, recovery transmissions, mobility, and multi-device contention. This paper introduces Deterministic-5G (Det-5G), a unified radio resource allocation framework for industrial closed-loop control. Det-5G treats the complete bidirectional control cycle as the scheduling object and combines coordinated downlink/uplink allocation, adaptive bundled transmissions, group-oriented downlink communication, and optimized multi-user uplink scheduling over 5G air-interface. Its performance is evaluated through a combination of closed-form analysis and Monte Carlo scheduling experiments, with comparisons against conventional dynamic grant-based scheduling, semi-persistent scheduling/configured grant operation, and fixed proactive repetition. The evaluation shows that Det-5G improves predictability of cycle completion, maintains the target reliability under changing link conditions, and scales more effectively to multi-device control than conventional reactive scheduling, while adapting radio resource use instead of continuously provisioning for the worst case as in fixed repetition. These characteristics make cycle-oriented scheduling a pragmatic solution for reducing the determinism gap that limits the use of 5G for closed-loop control in different verticals, especially as it evolves through 5G-Advanced toward 6G. Index Terms—5G, 6G, closed-loop control, determinism, Industry 5.0, private networks, resource allocation, scheduling, uRLLC.
Radio frame (10 ms)
1
0
3
2
5
4
7
6
8
9 Device K
Subframe (1 ms) 0
1
2
3
4
5
6
7
8
9
gNB
gNB 10
11
12
13
15 kHz (1 slot)
0.5 ms
Controller Single-user closed-loop control
Device
Controller Device 2
Multi-user closed-loop control
30 kHz (2 slots)
Device 1
0.25 ms
60 kHz (4 slots)
14-symbol TTI
7-symbol short TTI
slot (short TTI)
Mini-slot
Fig. 1. Illustration of scalable NR numerology and SBT/NSBT opportunities.
GALLOP targeted high-performance wireless closed-loop control for industrial IoT [8], [9]. These studies showed the importance of designing communication around periodic command/feedback exchange, tight timing, and reliability, but did not address cellular mechanisms such as 5G grant procedures or NR resource allocation. Within 5G, uRLLC research has extensively studied latency, reliability, retransmission, and scheduling. Prior work has examined factory-automation requirements [10], retransmissionaware allocation and communication-control co-design [11], [12], scheduler-induced burst errors for periodic traffic [13], and semi-persistent/configured-grant operation for time-sensitive traffic [14], [15]. More general industrial schedulers provide per-flow guarantees or joint uplink/downlink allocation [16], [17], but typically retain packet- or flow-level scheduling with reactive recovery. The gap addressed here is the lack of a unified cycle-oriented treatment that combines proactive reliability, cycle-level recovery, group downlink delivery, and scalable multi-device resource allocation within 5G NR.
Fig. 2. Single-user and multi-user closed-loop control scenarios for Det-5G.
mechanisms reduce recurring control overhead, but they do not intrinsically bind the two directions into one command/feedback transaction. When HARQ is invoked, cycle completion depends on random decoding outcomes. Fixed proactive repetition removes this reactive timing uncertainty, but a small repetition factor becomes unreliable as the channel degrades whereas a worst-case factor consumes the same resources in every cycle. Det-5G is designed to jointly address cycle predictability, target reliability, and adaptive resource use. IV. D ET-5G: D ESIGN AND P ROTOCOL O PERATION A. System model and design principles
Fig. 2 shows the two scenarios considered by Det-5G: singleuser closed-loop control and multi-user closed-loop control. The controller is connected to the gNB through a local wired/edge path, while the controlled device(s) use the NR air interface. The controller may execute at an industrial edge platform or elsewhere in the local network; the scope here is the radiointerface part of the control cycle. Devices are assumed to be connected before cyclic operation starts, since connection setup does not belong to the recurring control deadline. Det-5G follows four design principles. First, downlink/uplink co-design recognizes that command and feedback are causally coupled and should not be scheduled as unrelated packets. Second, self-contained cyclic allocation reserves the control and data resources needed to complete one cycle. Third, bundled III. 5G NR AND R ESOURCE A LLOCATION P RELIMINARIES low-latency transmissions place reliability resources before the A. NR time structure deadline, with a bundle length that adapts to link conditions. 5G NR supports scalable OFDM numerology with SCS Fourth, multi-user optimization uses group communication in ∆f = 15 × 2µ kHz. With normal cyclic prefix, a slot contains downlink and time-frequency packing in uplink so that cycle 14 OFDM symbols and its duration decreases as SCS increases. time does not grow linearly with the number of controlled NR also permits shorter transmissions over fewer symbols, devices. including 2-, 4-, and 7-symbol transmission intervals, allowing B. Single-user closed-loop control non-slot-based placement at finer time granularity [18]. We use A typical cycle consists of a controller command followed slot-based transmission (SBT) for allocation over a complete selected transmission interval and non-slot-based transmission by sensing or actuation feedback from the device; the same (NSBT) for flexible symbol-level placement. Fig. 1 illustrates framework applies if the application reverses the direction order. The fundamental Det-5G scheduling object is the self-contained these opportunities. transmission in Fig. 3. It begins with a joint downlink/uplink B. Conventional scheduling for cyclic control resource allocation that identifies resources for the downlink In conventional dynamic operation, downlink and uplink are command, the subsequent uplink response, and the final cyclescheduled independently; uplink additionally requires schedul- level acknowledgement/control. The downlink may contain one ing request/grant signaling before data transmission. HARQ transmission or a bundle of repeated transmissions. After the improves reliability reactively through decoding feedback and downlink phase, the device transmits its uplink response, again retransmission. For periodic traffic, semi-persistent scheduling as a single transmission or a bundle. A block acknowledgement (SPS) can reserve recurring downlink resources and configured is issued after the uplink bundle rather than after each individual grant (CG) scheduling can preconfigure uplink resources; we element. Thus, the resources required for one complete control refer to their combined periodic operation as SPS/CG. These transaction are determined together.
Joint downlink/uplink grant
Downlink transmission (bundle length = 2)
Block ACK
Downlink or uplink transmissions
Controller Bundle length update required Yes
gNB
Trigger bundle length update procedure
Keep the default or previously allocated bundle length
Find new bundle length based on transmission type
Device
Self-contained transmission (one cycle of control application)
No
Uplink transmission (bundle length = 3)
Fig. 3. Self-contained cyclic transmission for single-user Det-5G.
Fig. 4. Single-user bundle-length update procedure in Det-5G.
Algorithm 1: Bundle-length update (single-user) Input: ΓTh , RSSIAvg , IM , Bmax , previous/default B Set RSSIsingle ← RSSIAvg and obtain Γsingle ; Set Γbundle ← Γsingle and B ← Bprev ; while Γbundle < ΓTh + IM and B < Bmax do B ← B + 1; Update Γbundle using the additional bundle element; Output: B
A self-contained cycle can be realized entirely with SBTs or with a combination of SBTs and NSBTs. Once cyclic control starts, the allocation can operate in two ways. In reserved operation, the downlink/uplink resources of the self-contained transmission repeat periodically, eliminating recurring grant exchange for stable traffic. In cycle-by-cycle operation, a joint allocation is provided for the next cycle or group of cycles. The latter allows bundle length and resource placement to piggybacked on the uplink response. At the start of cyclic track mobility, interference, or changing link conditions. This operation, the gNB can use a default bundle length derived distinction is important: periodic reservation reduces signal- from connection-quality measurements. Thereafter, the update ing, whereas cycle-by-cycle operation maximizes adaptability trigger in (1) prevents unnecessary bundle changes when the link remains stable while reacting to sustained degradation or without abandoning the self-contained-cycle structure. The complete cycle may itself be repeated when the rapidly varying reception. Downlink and uplink bundles are transaction fails. Failure occurs if the downlink command adapted separately because the two directions may experience is not decoded and therefore no valid uplink response is different interference and link budgets. For a candidate bundle, let Γsingle denote the quality of generated, or if the uplink bundle cannot be decoded. Instead of sending an ordinary block acknowledgement, the gNB can a single transmission and Γgain (i) the reliability/coding gain issue a new joint allocation for repeating the self-contained contributed by the ith bundle element. The effective quality is cycle. Bundling is also applicable to control information when represented as Γ(i) = Γsingle + Γgain (i). Algorithm 1 selects the smallest bundle that satisfies the receiver target plus an additional robustness is required. Det-5G further uses bundled transmissions as a proactive interference margin while respecting Bmax . This preserves the HARQ mechanism. Reliability resources are placed before original Det-5G principle: reliability is adapted before the cycle waiting for a conventional HARQ feedback/recovery round. rather than discovered through reactive retransmission delay. A bundle may repeat coded data, may distribute data and D. Multi-user closed-loop control additional FEC across its elements, or may use elements Directly repeating the single-user procedure for every device with different FEC contributions for the same payload. The key property is not a particular coding realization; it is would make both command and feedback airtime scale with that the scheduler knows the reliability expenditure and the group size. Multi-user Det-5G therefore modifies the selfcontained cycle in two ways: (i) a group downlink (G-downlink) corresponding cycle budget before transmission begins. delivers the controller command to the complete group and (ii) C. Single-user bundle-length adaptation a multi-user bundled uplink packs feedback transmissions from A large bundle improves reliability but consumes time- multiple devices into the same time-frequency region. Fig. 5 frequency resources. Det-5G therefore adapts downlink and shows the resulting cycle. The joint allocation precedes the uplink bundle lengths using recent received-signal statistics. Let G-downlink bundle, followed by the packed uplink bundle and RSSIAvg and RSSIVar denote the mean and variance observed a block acknowledgement after successful reception from the for the most recent bundle, RSSIK the moving average over group. the previous K bundles, and RSSITh and RSSIVarTh design A logical group identifier enables the devices belonging to thresholds. An update is triggered when the control group to decode the common downlink transmission. Groups may be predefined in static or quasi-static installations RSSIAvg < RSSITh ∨ RSSIAvg < RSSIK ∨ (1) or created dynamically as link conditions, mobility, or control RSSIVar > RSSIVarTh . participation change. Dynamic grouping is also useful for The closed-loop exchange provides fresh measurements con- recovery: after a self-contained cycle, devices that did not tinuously; for example, downlink quality information can be complete the transaction can be regrouped and assigned a
Group information
Joint downlink/uplink grant
G-downlink transmission (bundle length = 2)
Block ACK
Controller Multi-user bundled uplink transmission (group bundle length = 3) gNB
Frequency
4
Devices
1 2 3
3 4 2
2 3 1
1
Self-contained transmission (one cycle of control application)
Time
Fig. 5. Self-contained cyclic transmission for multi-user Det-5G.
Algorithm 2: Bundle-length update for G-downlink Input: Group M and inputs of Algorithm 1 Find w ∈ M with the worst received link quality; Run Algorithm 1 for device w; Use the resulting BGDL for the G-downlink bundle; subsequent joint allocation instead of forcing successful devices to repeat unnecessarily. E. G-downlink optimization and group recovery
Algorithm 3: Multi-user uplink optimization (SBTs) Input: M, NRB , MCS table Q and inputs of Algorithm 1 min Find required bundle length Bm for every device using Algorithm 1; Sort devices in descending order of received SNR; set GBL ← 0; min while some device has fewer than Bm scheduled transmissions do GBL ← GBL + 1; reset available RBs; foreach eligible device m do Map SNRm to Q and determine its RB demand; Allocate RBs if capacity remains; update the scheduled bundle of m; Output: GBL and the SBT allocation Algorithm 4: Multi-user uplink optimization (NSBTs) Input: Inputs of Algorithm 3 Find Bnmin , RB demand, and transmission duration Tn for each device; Define TTxn as the set of outstanding transmissions; while TTxn ̸= ∅ do Select the outstanding transmission with largest Tn ; Allocate its required time-frequency resources; Schedule transmissions of other devices in residual feasible opportunities; Update TTxn ; Output: Bundled NSBT allocation
Since every member must decode the common command, the G-downlink bundle is governed by the limiting device. For group M, the gNB maintains received-quality statistics for every device and applies the single-user adaptation to the device w with the worst current link. Algorithm 2 therefore turns the individual bundle procedure into a group reliability rule. This worst-device rule guarantees a common downlink reliability target, but it also exposes an important grouping trade-off: repeatedly placing a persistently weak device in a strong group can increase the G-downlink bundle for all members. Dynamic grouping can therefore cluster devices with an MCS and hence to the RB demand needed for its payload, comparable link conditions or isolate a temporarily degraded min headers, and redundancy. Let Bm denote the bundle required device for a subsequent cycle. by device m, Γm its received quality, and NRB the RB pool The block acknowledgement also operates at group-cycle in an SBT. The objective is to minimize the group bundle level. A successful multi-user cycle requires the scheduled length GBL while meeting every device’s bundle and quality uplink response from every participating device; otherwise, requirements: the gNB identifies the incomplete subset and can issue a new joint allocation only for that subset. Consequently, successful min min GBL s.t. Bm ≥ Bm , Γm ≥ ΓTh , ∀m ∈ M. (2) devices need not repeat their command/feedback exchange solely because another member failed. The same principle can be used with periodic reservation: the regular group Algorithm 3 first determines the individual bundle requirements schedule remains reserved, while exceptional recovery cycles and then packs transmissions into the available RB pool. The are inserted for failed members. This keeps recovery aligned group bundle length increases only when another SBT is with the control transaction rather than with individual packet required. For NSBT operation, the allocation is refined to symbol-level retransmissions. time-frequency regions rather than forcing every transmission F. Multi-user bundled uplink optimization to occupy the full SBT duration. A device with a high-rate For the uplink, each device can require a different number MCS may therefore finish in only a few symbols while a of repetitions and a different number of resource blocks (RBs) weaker device occupies a longer region. Let Tn denote its rebecause its link quality and MCS differ. Det-5G therefore quired symbol duration. The scheduler minimizes the occupied optimizes the bundled uplink in both time and frequency. Two bundled NSBT length SL subject to the same reliability/quality cases are distinguished according to whether the self-contained constraints. The longest transmission is scheduled first and transmission uses SBTs or NSBTs. residual time-frequency opportunities are filled by other devices, For SBT operation, an RB is the scheduling unit in frequency as summarized in Algorithm 4. This allows short transmissions and the selected SBT interval is the scheduling unit in time. to use otherwise idle symbols and is particularly effective for Each device is first mapped from its measured link quality to heterogeneous links and payloads.
V. P ERFORMANCE E VALUATION
TABLE I E VALUATION PARAMETERS
The evaluation combines closed-form timing/reliability analysis with a customized 5G air-interface scheduler and Parameter Setting Monte Carlo simulations. Figs. 6, 7, and 9 are analytical Carrier / bandwidth 3.8 GHz / 40 MHz under the stated link/recovery model; hence they have no Message size 64 bytes per direction Main SCS / SBT 30 kHz / 7 symbols (0.25 ms) sampling uncertainty. Fig. 8 uses 10,000 independent scheduler Available RBs 106 realizations per device-count point and reports distribution-free Link state probabilities 0.50 / 0.35 / 0.15 95% order-statistic confidence intervals for the 99th percentile. BLER values 10−3 /10−2 /10−1 MCS selection CQI-driven, link-adaptive Fig. 10 uses 30 independent runs of 100,000 cycles per speed Main deadline / cycle target 4 ms / 2 × 10−5 and reports 95% confidence intervals. Stringent stress point (Fig. 9) 1 ms / 10−6 We use 5G-NR configurations and a periodic industrial Recovery interval 4 transmission quanta Controlled devices / mobility 1–20 / 0–10 m/s motion-control traffic profile with 64-byte messages per direction over a 3.8 GHz1 , 40 MHz carrier. The main configuration 10 uses 30 kHz SCS, a 7-symbol SBT of 0.25 ms, and 106 RBs. Average cycle time 3.5 CQI/MCS is selected according to current link condition for Cycle failure probability Target 2 × 10 10 3.0 resource sizing. BLER values 10−3 , 10−2 , and 10−1 represent favorable, moderate, and challenging states with probabilities 2.5 0.50, 0.35, and 0.15. A per-direction residual target 10−5 gives 10 2.0 an approximate cycle target 2 × 10−5 . Dynamic and SPS/CG 1.5 baselines use reactive HARQ; fixed K = 2 and K = 5 isolate 10 1.0 proactive repetition. The main recovery increment is 1 ms (four 0.5 SBT quanta). Table I summarizes the assumptions [19], [20]. 10 0.0 For the analytical evaluation, the bundle assigned to link G mic /CG K=2 K=5 Det-5 SPS Dyna Fixed Fixed state s is the smallest integer whose residual per-direction error probability meets ϵdir . Under the independent repetition model Fig. 6. Average cycle time and cycle failure probability under a common used for the comparison, 4 ms deadline. Values are obtained analytically; the horizontal line is the cycle-failure target. ln ϵdir Bs = , (3) ln ps eventual cycle completion time under the link-state mixture; the where ps is the BLER of state s. This mapping is an evaluation right axis is the probability that the complete downlink/uplink model for quantifying the proactive reliability budget; it is transaction misses the deadline or is not delivered successfully. consistent with the design objective of Algorithm 1, while For reactive HARQ, the aggregate number of DL and UL the protocol itself may obtain the bundle from measured link failures follows a negative-binomial distribution. For fixed quality and coding gain. repetition, the residual cycle failure at BLER p is 1−(1−pK )2 ; For the reactive HARQ baselines, we model the numbers of failed DL and UL transmissions before successful delivery as Det-5G uses the corresponding state-dependent bundle B. Dynamic scheduling averages 1.54 ms but has cycle failure independent geometric random variables. From their aggregate recovery process, we derive the final expressions used for the 5.6×10−4 ; SPS/CG averages 1.04 ms but remains at 6.9×10−5 . cycle deadline-failure probability and expected completion time Fixed K = 2 is short (1.75 ms) but unreliable at 3.1 × 10−3 . as " # Fixed K = 5 reaches 3.0 × 10−6 but fixes every cycle at nD X X 2 n 3.25 ms. Det-5G reaches 4.7 × 10−6 with 2.15 ms average Pfail,HARQ (D) = πs 1 − (n + 1)(1 − ps ) ps , s n=0 (4) cycle time. Thus average latency alone is insufficient: relative X 2ps to fixed K = 5, Det-5G lowers average cycle time by about T HARQ = πs T0 + TR . 1 − p 34% and average proactive repetition occupancy by 44% while s s Here nD = ⌊(D − T0 )/TR ⌋ is the number of recovery satisfying the target. −5
−3
−4
−5
Cycle failure probability
Average cycle time (ms)
−2
−6
opportunities that fit before deadline D, T0 is the no-recovery cycle time, TR is the recovery interval, ps is the BLER of state s, and πs is its occurrence probability. For the singleuser P Det-5G comparison, the corresponding average is T Det = s πs (C + 2Bs )Tq , where C = 3 fixed allocation/control quanta and Tq is the transmission-quantum duration. A. Joint cycle time and deadline reliability Fig. 6 evaluates the central trade-off under the common 4 ms application deadline. The left axis is the exact expected 1 The 3.8-4.2 GHz band is available for private 5G deployments in the UK.
B. Retransmission-induced cycle time variation Fig. 7 isolates reactive timing uncertainty at fixed BLER. The 99.9th–0.1th percentile spread is calculated directly from the exact negative-binomial distribution of aggregate DL/UL HARQ failures, rather than estimated from finite trials. HARQbased scheduling develops 1–3 ms of cycle time variation across the evaluated BLER range. Det-5G and fixed proactive schedules have zero reactive variation once a cycle allocation is selected; link adaptation may change the predetermined budget between cycles but does not append an unplanned HARQ tail to the allocated cycle.
2.5
HARQ-based scheduling Det-5G / fixed proactive schedule
2.0 1.5 1.0 0.5 0.0 10−3
10−2
10−1
BLER
99th-percentile cycle time (ms)
Cycle time variation (ms)
3.0
5 4 3 2
Conventional grant-based + HARQ Periodic reservation (SPS/CG) + HARQ Det-5G (SBT) Det-5G (NSBT)
1 0 1
4
8
12
16
20
Controlled devices per cycle
Fig. 7. Analytical retransmission-induced cycle time variation, measured as the 99.9th–0.1th percentile spread at fixed BLER.
C. Multi-user cyclic scalability
Fig. 8. 99th-percentile cycle time with multi-user frequency multiplexing enabled for all schemes. Error bars are 95% order-statistic confidence intervals.
Cycle time (ms)
Dynamic + HARQ 25 Fig. 8 removes the weak sequential baseline used in an SPS/CG + HARQ earlier formulation. The conventional dynamic and SPS/CG Fixed repetition, K=7 20 Det-5G schedulers are allowed to frequency-multiplex independent 1 ms stress deadline 15 UE transmissions within the same SBT using the same 10610 RB pool and CQI-dependent RB demand as Det-5G. They retain per-UE DL/UL transactions and reactive HARQ, whereas 5 Det-5G uses one G-downlink and proactively packs bundled 0 15 kHz 30 kHz 30 kHz 60 kHz 60 kHz uplink transmissions. The figure therefore compares the 99th14 sym 14 sym 7 sym 7 sym 2 sym 5G air-interface configuration percentile cycle time rather than mean latency, since the latter favors reactive schemes when no recovery is needed. Resource −6 demand is computed consistently for all schemes as NRB = Fig. 9. Analytical cycle time across 5G air-interface configurations at a 10 cycle-failure target. The 1 ms line is a stringent full-cycle stress deadline; ⌈L/(ηCQI NRE )⌉, where L is the transmitted bit count, ηCQI fixed K = 7 coincides with the Det-5G worst-state timing bound. is the spectral efficiency associated with the sampled CQI, and NRE is the usable resource-element count for the selected fixed K = 7 therefore has the same worst-state timing bound transmission duration. In the NSBT model, CQI values 11–15, but repeats seven times in every state. At 60 kHz/7 symbols, 7–10, and 1–6 use 2-, 4-, and 7-symbol transmission durations, the resulting cycle times remain 3.25 ms (dynamic), 3.0 ms (SPS/CG), and 2.125 ms (Det-5G/fixed K = 7). With a 60 kHz respectively. The fairer comparison substantially reduces the previously 2-symbol NSBT opportunity they fall to 0.93, 0.86, and 0.61 ms, observed gain and changes the interpretation. At one or two respectively. Hence the strict 1 ms full-cycle stress point is devices the conventional baselines have lower tail cycle time. reached only with sufficiently fine transmission granularity in As group size grows, however, the probability that at least the evaluated set; Det-5G provides the largest timing margin one independent transaction requires recovery increases. At while avoiding persistent worst-case repetition. 16 devices, the 99th-percentile cycle times are 5.25 ms for dynamic scheduling, 4.75 ms for SPS/CG, 3.75 ms for Det-5G E. Mobility and bundle update overhead SBT, and 3.43 ms for Det-5G NSBT. Thus NSBT lowers the Mobility does not imply that Det-5G must change its bundle 99th-percentile cycle time by about 35% and 28% relative to every control cycle. Link quality is measured continuously, the dynamic and SPS/CG baselines, respectively. Error bars while the bundle changes only after a persistent state transhow 95% order-statistic confidence intervals from 10,000 sition. We use Tc ≈ 0.423/fD at 3.8 GHz and a two-cycle independent realizations per point. persistence rule. The coherence time is mapped to a per-cycle state-transition probability q = 1 − exp(−Tcp /Tc ), where D. Stringent latency/reliability stress test Tcp = 1 ms is the reference control period used in this The 5G-ACIA motion-control SLS example lists 1 ms end- sensitivity study. Fig. 10 reports the mean update frequency to-end latency, 1 ms transfer interval, zero survival time, and from 30 independent runs of 100,000 cycles per speed. Bundle 99.9999% communication-service availability [4]. Availability updates occur in about 2.81%, 7.48%, 11.10%, and 16.70% of is a service-level quantity and is not identical to the per-cycle cycles at 1, 3, 5, and 10 m/s; all nonzero-speed 95% confidence radio failure probability used here, so we do not convert one half-widths are below 0.04 percentage points. directly into the other. Instead, Fig. 9 adds an intentionally Relative to cycle-by-cycle Det-5G, event-triggered reserved stringent radio-side stress point: a 1 ms full-cycle deadline and operation therefore reduces bundle update actions by approx10−6 cycle-failure target. imately 97.2%, 92.5%, 88.9%, and 83.3% at 1, 3, 5, and Meeting 10−6 requires five aggregate HARQ recovery 10 m/s. SPS/CG can require less routine reconfiguration, but opportunities for the reactive baselines under the link-state recovery still introduces timing uncertainty; fixed worst-case mixture. Det-5G requires state-dependent bundles B = 3, 4, 7; repetition requires little adaptation but pays continuous resource
Cycles with bundle update (%)
R EFERENCES 15
10
5
0 0.0 0.5 1.0
3.0
5.0
10.0
Device speed (m/s)
Fig. 10. Event-triggered Det-5G bundle update frequency versus device speed.
cost. Reserved Det-5G occupies the middle ground: reliability resources change only when the link state persistently changes, while the allocated cycle remains predetermined. VI. C ONCLUDING R EMARKS This paper introduced Det-5G, a cycle-oriented radio resource allocation solution for deterministic industrial closedloop control over 5G. Det-5G treats the complete bidirectional control cycle as the scheduling unit and combines coordinated downlink/uplink allocation, proactive link-adaptive bundled transmissions, cycle-level recovery, group-oriented downlink communication, and SBT/NSBT-based multi-user uplink optimization. The resulting design aims to bound control-cycle completion while maintaining reliability and adapting radio resource use to changing link conditions and multi-device operation. Under the main 4 ms operating point, dynamic and SPS/CG scheduling achieve shorter average cycle times but exceed the cycle-failure target, whereas Det-5G satisfies the target with a 2.15 ms average cycle and uses 44% less proactive repetition occupancy than fixed K = 5. Analytical characterization further shows that reactive HARQ introduces 1–3 ms cycle time variation. With multi-user frequency multiplexing enabled for the conventional baselines, Det-5G NSBT provides a 28– 35% reduction in 99th-percentile cycle time at 16 devices. The stringent stress study further shows that a 1 ms full-cycle/10−6 radio target requires sufficiently fine transmission granularity, with Det-5G providing the largest timing margin among the evaluated schemes. Mobility adaptation remains event-triggered, requiring bundle updates in only about 7.5% of cycles at 3 m/s. These results position Det-5G as a system-level way to address an unresolved industrial 5G adoption barrier as networks evolve through 5G-Advanced, while using existing NR mechanisms rather than depending on a new 5G-Advanced primitive. The framework is also complementary to 5G/TSN convergence: synchronization, QoS coordination, and deterministic Ethernet integration do not remove radio-side timing uncertainty. Bounding the complete wireless control transaction can therefore help extend deterministic behavior across converged wired/wireless systems and support progression from connectivity-oriented private 5G toward closed-loop automation, coordinated robotics, Physical AI, and ultimately 6G cyber-physical communication.
[1] A. Aijaz, “Private 5G: The Future of Industrial Wireless,” IEEE Ind. Electron. Mag., vol. 14, no. 4, pp. 136–145, 2020. [2] 3GPP, “5G for Industry 4.0,” 2020, overview of Release-16 TSN/TSC, NPN and 5G-LAN support. [Online]. Available: https: //www.3gpp.org/technologies/tsn-v-lan [3] ——, “Service Requirements for Cyber-Physical Control Applications in Vertical Domains,” 3rd Generation Partnership Project (3GPP), TS 22.104, 2026, specification under change control. [4] 5G Alliance for Connected Industries and Automation (5G-ACIA), “Service-Level Specifications (SLSs) for 5G Technology-Enabled Connected Industries,” 5G-ACIA / ZVEI, Tech. Rep., Sep. 2021, motioncontrol example: 1 ms end-to-end latency, 1 ms transfer interval, zero survival time, and 99.9999 percent communication-service availability. [Online]. Available: https://5g-acia.org/download/18765/?version=A4 [5] ——, “Performance Testing of 5G Systems for Industrial Automation,” 2021, white paper. [Online]. Available: https://5g-acia.org/whitepapers/ performance-testing-of-5g-systems-for-industrial-automation/ [6] 3GPP, “Release 20,” 2026, 5G-Advanced technical specifications and early 6G studies. [Online]. Available: https://www.3gpp.org/ specifications-technologies/releases/release-20 [7] P. Li, X. Lin, and A. Aijaz, “Rethinking Networked Robotics in the 6G Era With Generative AI-in-the-Loop,” IEEE Communications Standards Magazine, vol. 10, no. 2, pp. 287–295, 2026. [8] A. Aijaz, “ENCLOSE: An Enhanced Wireless Interface for Communication in Factory Automation Networks,” IEEE Trans. Ind. Informat., vol. 14, no. 12, pp. 5346–5358, 2018. [9] A. Aijaz and A. Stanoev, “Closing the Loop: A High-Performance Connectivity Solution for Realizing Wireless Closed-Loop Control in Industrial IoT Applications,” IEEE Internet Things J., vol. 8, no. 15, pp. 11 860–11 876, 2021. [10] S. A. Ashraf et al., “Ultra-reliable and Low-latency Communication for Wireless Factory Automation: From LTE to 5G,” in IEEE ETFA, 2016, pp. 1–8. [11] S. E. Elayoubi, P. Brown, M. Deghel, and A. Galindo-Serrano, “Radio Resource Allocation and Retransmission Schemes for uRLLC Over 5G Networks,” IEEE J. Sel. Areas Commun., vol. 37, no. 4, pp. 896–904, 2019. [12] B. Chang et al., “Optimizing Resource Allocation in uRLLC for RealTime Wireless Control Systems,” IEEE Trans. Veh.Tech., vol. 68, no. 9, pp. 8916–8927, 2019. [13] J. S. Demel, C. Bockelmann, and A. Dekorsy, “Burst Error Analysis of Scheduling Algorithms for 5G NR uRLLC Periodic Deterministic Communication,” in 2020 IEEE 91st Vehicular Technology Conference (VTC2020-Spring), 2020, pp. 1–5. [14] N. Jiang, A. Aijaz, and Y. Jin, “Recursive Periodicity Shifting for SemiPersistent Scheduling of Time-Sensitive Communication in 5G,” in IEEE Global Communications Conference (GLOBECOM), 2021, pp. 01–06. [15] T. Zhang, X. S. Hu, and S. Han, “Contention-Free Configured Grant Scheduling for 5G uRLLC Traffic,” in ACM/IEEE Design Automation Conference (DAC), 2023, pp. 1–6. [16] T. Zhang, J. Wang, X. S. Hu, and S. Han, “Real-Time Flow Scheduling in Industrial 5G New Radio,” in IEEE Real-Time Systems Symposium (RTSS), 2023, pp. 371–384. [17] L. Kleinberger, M. Gundall, and H. D. Schotten, “FLEX: Joint UL/DL and QoS-Aware Scheduling for Dynamic TDD in Industrial 5G and Beyond,” in IEEE International Conference on Communications (ICC), 2026, preprint arXiv:2603.20971. [18] 3GPP, “NR; Physical Layer Procedures for Data,” 3rd Generation Partnership Project (3GPP), TS 38.214, Jun. 2026, Release 19, v19.4.0. [19] 5G Alliance for Connected Industries and Automation (5G-ACIA), “A 5G Traffic Model for Industrial Use Cases,” 2019, white paper; motion-control traffic includes small periodic messages and 1 ms sensor transfer intervals. [Online]. Available: https: //5g-acia.org/whitepapers/a-5g-traffic-model-for-industrial-use-cases/ [20] 3GPP, “NR; Base Station (BS) Radio Transmission and Reception,” 3rd Generation Partnership Project (3GPP), TS 38.104, 2026, release 19.