SSF: 5G Core Sync for Satellite Handover

• SSF is a core-network synchronization proxy that enables parallel handover in LEO satellite systems by decoupling RAN execution from core signaling.
• It pre-stores critical signaling information and triggers the path switch at a predetermined instant, reducing handover latency from 250 ms to 8.8 ms.
• SSF integrates a hybrid ML-based signal prediction and optimized satellite selection, achieving significant improvements in latency and stability without modifying existing core functions.

Satellite Synchronized Function (SSF) is a novel 5G core network function proposed in PHandover for low-latency handover in regenerative Low Earth Orbit (LEO) satellite networks. Its defining role is to decouple handover execution in the RAN from signaling interaction with the core network while remaining fully compliant with the standard 5G core network. In the PHandover design, SSF stores pre-handover signaling, including PathSwitchRequest-type information, and triggers the standard path switch at a planned instant so that the access-network switch and the downlink path switch proceed in parallel rather than serially. The function is introduced to remove the dominant handover delay associated with the long signaling path between satellite RAN and the ground-based core network, which may traverse multiple ISLs and a satellite-ground hop (Wu et al., 10 Jul 2025).

1. Functional definition and architectural position

Within PHandover, SSF is the key core-network-side enabler of a plan-based, parallel handover mechanism for mobile satellite networks. The core design premise is that, in regenerative mode, satellites act as S-gNBs, and the bottleneck in handover is often not the radio-side switch itself but the latency of access-to-core signaling. Standard 5G NTN handover is therefore too slow when it requires sequential interaction between the access and core networks.

SSF is explicitly positioned as a novel network function that can be added to the 5G core network with no modification to existing core NFs. It is not a new access node, satellite, or user-plane function. Rather, it is a core-network synchronization and signaling proxy function that enables parallelized access/core handover timing and avoids the need for real-time handover signaling from the access side at the instant of switch (Wu et al., 10 Jul 2025).

A central compatibility mechanism is that SSF disguises itself as the source S-gNB in the handover signaling flow. Existing core entities such as AMF and UPF therefore see an ordinary-looking handover signaling sequence, even though the timing is prearranged and synchronized. This preserves standard 5G core procedures while changing the temporal relation between access-network execution and core-network path switching.

2. Procedural logic of parallel handover

The SSF-based procedure splits handover into a preparation phase and an execution phase. This separation is the architectural basis for parallelization.

During handover preparation, the source S-gNB performs the planning work in advance:

1. The source S-gNB predicts future UE access satellites and the handover moment using the trajectory/prediction engine.
2. It informs the target S-gNB in advance.
3. The target S-gNB prepares resources, such as channel reservation, and confirms the pre-handover.
4. The pre-handover result is sent to SSF, including the UE’s NGAP ID, target S-gNB identity, and the scheduled path-switch timing.

During handover execution, access and core are synchronized rather than serialized. On the access-network side, the source S-gNB notifies the UE of the handover decision; the UE disconnects from the source and establishes a new RRC connection to the target S-gNB. Simultaneously, the source S-gNB sends synchronized data and sequence number information to the target S-gNB, and then the source is released. On the core-network side, SSF initiates the path-switch request at the planned time, and the normal UPF-mediated downlink path switch proceeds (Wu et al., 10 Jul 2025).

The preparation logic is designed to mimic standard 5G procedures closely, except that the target S-gNB assigns the NGAP ID earlier than in baseline 5G so that SSF has all required information before execution. This earlier assignment is necessary because SSF must hold all signaling state required to trigger the path switch without waiting for a fresh access-side report at execution time.

3. Contrast with standard NTN and conditional handover

The main procedural distinction between SSF-enabled handover and standard NTN/5G handover is temporal ordering. In standard 5G NTN handover, the source gNB or S-gNB generally completes the access-side switch first and then informs the core network, which in turn performs the downlink path switch. The resulting process is sequential and core-dependent.

Even conditional handover (CHO), which 3GPP recommends as a baseline for NTN, does not remove the core-network synchronization delay. CHO mainly pre-establishes Xn connections to multiple candidates and pre-reserves resources, but it still relies on gNB-to-core interaction during downlink switching. SSF changes that dependency by moving the planning of core-network signaling earlier, before execution, and then issuing the necessary signaling at the correct moment.

This implies a different division of labor. Standard schemes are measurement-driven and sequential, whereas SSF supports a plan-based, parallel handover. A plausible implication is that the principal gain does not arise from a new radio procedure, but from moving core-network synchronization from the critical path to a precomputed control path. The paper characterizes this as decoupling handover execution in the RAN from signaling interaction with the core network while keeping the core network itself unchanged (Wu et al., 10 Jul 2025).

4. Prediction, timing control, and access-satellite selection

SSF does not operate in isolation. It is coupled to a machine learning-based signal strength prediction model and a handover scheduling algorithm, because the path switch must be triggered at a precise planned time.

The prediction model is hybrid: it combines a 3GPP-style satellite-to-ground channel model with data-driven learning. The input features are

$[\theta, h, \theta_0, s_0],$

where $\theta$ is the elevation angle, $h$ is the satellite height, and $\theta_0, s_0$ are an earlier measurement’s elevation angle and signal strength. The paper defines the path-loss decomposition as

$PL = PL_b + PL_g + PL_s + PL_e,$

with

$PL_b = FSPL(d,f) + SF + CL(\alpha,f),$

$$FSPL(d,f) = 32.45 + 20 \log_{10}(f) + 20 \log_{10}(d),$$

$SF \sim N(0,\sigma_{SF}^2),$

$PL_g = \frac{A_{zenith}(f)}{\sin(\alpha)},$

and the predicted signal strength model

$$s_t^{\text{model}} = s_0 - 20\log_{10}\!\left(\frac{d}{d_0}\right) - \frac{A_{zenith}(f)}{\sin(\alpha)} + \frac{A_{zenith}(f)}{\sin(\alpha_0)}.$$

The model output is then fed into the learning component, which helps the S-gNB forecast future access satellites and decide when SSF should schedule the path switch (Wu et al., 10 Jul 2025).

The handover scheduling algorithm performs periodic prediction every $\theta$0, using a table $\theta$1 of access-satellite assignments. At time $\theta$2, the S-gNB plans for $\theta$3 to $\theta$4. It retrieves the access satellites at the beginning of the interval and predicts the access satellites at the end of the interval. If a change is detected, it uses binary search to refine the handover trigger time $\theta$5, iteratively halving the timing error. The paper sets

$\theta$6

and states that with 9 binary-search iterations, the system can achieve 10 ms timing precision.

To reduce computation, the design adds a fast access satellite selection procedure. It first filters UEs based on access strategy. Under a consistent access strategy, the UE stays with the current satellite until it leaves coverage; under a flexible access strategy, the UE may handover earlier for better quality. It then restricts candidate satellites to those near the current satellite based on orbital distribution and block partitioning. The paper further describes an access-satellite selection optimization that constrains handovers to same-direction satellites. Contrary-direction satellite switches produce a sudden latency increase of 30–40 ms, while similar-direction handovers can reduce inter-satellite propagation delay by more than 200 ms in the best case.

5. Measured performance and empirical characteristics

The PHandover experiments identify latency as the principal benefit of SSF-enabled parallel handover. The proposed scheme achieves an average handover latency of 8.8 ms, compared with 250 ms for standard NTN handover, corresponding to about a 21× reduction. It is also lower than two other optimized schemes (Wu et al., 10 Jul 2025).

Scheme	Average handover latency
Proposed scheme	8.8 ms
Standard NTN handover	250 ms
NTN-GS	153 ms
NTN-SMN	158.5 ms

The same study reports that omitting the access-satellite direction optimization increases latency by about 6.1×, indicating that SSF’s timing advantage is amplified by smarter target selection. Between the flexible and consistent access strategies, the flexible strategy is about 10 ms faster on average but causes more frequent handovers.

Stability metrics show similar trends. The handover failure rate is around 0.3‰ for the proposed scheme, whereas competing methods are between 1.5‰ and 2.5‰ in Starlink. The packet loss rate is around 0.35‰ for the proposed scheme, versus 0.7‰ to 0.9‰ for baselines. At the user level, stalling time is reduced by 89% compared with 5G NTN handover, and for TCP flows stalling time is reduced by 33% (Wu et al., 10 Jul 2025).

The signal-strength prediction subsystem that feeds SSF scheduling is also evaluated. The hybrid predictor achieves median error 1.07 dB and 90th percentile error 2.63 dB. Compared with ITU/3GPP channel-model prediction, this reduces error from 1.6 dB / 4.5 dB to 1.07 dB / 2.63 dB, that is, about 33% and 43%. Compared with LSTM, the proposed method is very close: the LSTM median error is 1.06 dB and the LSTM 90th percentile error is 2.59 dB. The paper also states that the proposed predictor consistently outperforms the ITU baseline and is close to LSTM while requiring fewer historical inputs.

6. Overhead, deployment implications, and acronym ambiguity

The computational overhead per unit time of SSF-enabled scheduling is modeled as

$\theta$7

where $\theta$8 is the number of users, $\theta$9 the number of satellites, and $h$0 the fraction of satellites considered as candidates. The communication overhead is

$h$1

and the storage overhead is

$h$2

where $h$3 is handover frequency and $h$4 is signaling size. For a scenario with $h$5 users and 500 B signaling messages, the paper estimates about 1 MB/s additional network overhead and about 5 MB storage overhead, which it deems practical (Wu et al., 10 Jul 2025).

These numbers are consistent with the authors’ deployment-oriented claim that SSF can be inserted into the 5G core network without modifying existing core NFs. A plausible implication is that the main adoption barrier is not standards incompatibility at the AMF/UPF level, but the correctness of prediction, scheduling, and synchronized execution under operational load. The paper’s design overview reflects this by presenting SSF as one component of a broader PHandover architecture involving prediction, scheduling, and synchronized execution.

The acronym “SSF” is not unique across the literature. In mathematical physics, SSF commonly denotes the spectral shift function (Sambou, 2015), including recent lattice-operator work on engineering SSF singularities (Assal et al., 2024). In a different satellite-communications context, a 2026 study uses SSF in the sense of synchronized audiovisual reconstruction under semantic transmission (Liu et al., 11 Mar 2026). Within PHandover, however, SSF has a specific and narrower meaning: a core-network synchronization proxy for satellite handover that pre-stores signaling and issues the path-switch request at the planned time so that core-network downlink switching occurs in parallel with access-network RRC handover.