Master Thesis: Enhanced Holdover for High-Precision Timing Systems

Net Insight develops network timing and synchronization equipment for mobile networks. One of our products, Zyntai, provides

GNSS-independent time synchronization for 5G using time transfer. This thesis would be carried out together with our R&D team in

Stockholm and focus on precision timing systems.

5G networks need base stations to stay synchronized to within about a microsecond, measured as time error (TE). Too much drift

between nearby base stations causes interference and degrades the network. Base stations normally get this accuracy from GNSS, but

GNSS can be jammed, spoofed, or lost, and backup synchronization via time transfer isn't always available either. When that

happens, equipment falls back on holdover, using its own oscillator to keep time. Oscillators that seem accurate by human

standards drift into unacceptable TE within a few hours, and extending that window without resorting to an expensive, higher

quality oscillator is still an open industry problem. There is no fixed answer for how to approach it.

The problem is wide and can be tackled in different ways. A few directions worth looking into: Kalman filter based prediction of

oscillator drift, switching more smoothly between trusting the network and holding the current time, machine learning to let the

system learn an oscillator's behavior from data directly or combining information from more than one frequency source. Other

creative approaches are welcomed as well!

Ideally, the thesis will implement and evaluate multiple approaches. Which of these to focus on is something we would decide

together before starting the thesis work.

• A desk with the team at our Stockholm office, access to our lab equipment and test setups, the real gear we test our own

products with.

We’re looking for curious minds with a solid technical foundation and a passion for problem-solving! Check out this video where

previous master thesis students, now proud employees, share their experience of doing their master thesis with us.

Your education is ideally in electrical engineering, engineering physics, computer science, or an equivalent field, ideally with a

background in control theory, C++, Python, Linux, networking, and embedded systems. This project is suitable for 1-2 students.

Ready to turn your thesis into something truly impactful? Join us at Net Insight and let’s create the future of synchronization

together!

5G uses time division duplexing, alternating between sending and receiving on the same frequency in windows often below one

millisecond. This requires base stations to agree on time to within microseconds: if nearby base stations become sufficiently out

of sync, their transmissions interfere, degrading service or causing outages. A 5G base station is considered out of sync once its

time error (TE) exceeds 1 µs.

Base stations obtain this accuracy from GNSS satellites, which distribute reference time established by timing laboratories using

atomic clocks. This has long been the industry standard, but GNSS availability cannot be guaranteed: its signals can be jammed or

spoofed, making timing unreliable or unavailable in some regions.

When the primary reference is lost, equipment enters holdover, maintaining synchronization using its own local frequency source

until the reference returns. This source is typically an oscillator (TCXO/OCXO) rather than an atomic clock. Such oscillators are

highly accurate by human standards, drifting only milliseconds per year, but this small drift becomes significant at microsecond

scale: extrapolated to the 1 µs threshold, such an oscillator may keep equipment within limits for only around 3-5 hours.

Enhanced holdover studies how timing equipment can remain within required synchronization limits while its primary reference is

unavailable. The problem is not simply keeping a local clock running, but extending the time before accumulated error makes

equipment unusable for high-precision systems such as 5G base stations.

Timing equipment uses a physical oscillator as its frequency reference. Common types include quartz crystal oscillators,

temperature-compensated (TCXO) and oven-controlled (OCXO) variants for higher stability, and atomic references (rubidium, cesium)

where extreme stability is required. No oscillator runs at exactly its nominal frequency; the deviation is specified as accuracy

or stability, typically in parts per million (ppm) or parts per billion (ppb). Long-term average accuracy and short-term stability

(how much frequency wanders moment-to-moment) are related but distinct, and it is short-term stability that dominates holdover

performance.

A relative frequency error accumulates a time error over time when in holdover. An oscillator good enough to hold 1 µs of error

for three hours has a stability of approximately 1 µs ÷ 10,800 s ≈ 9.3×10⁻¹¹: about 1 part in 10 billion. The same stability

implies roughly 8 µs of drift per day and about 3 milliseconds per year, which illustrates the core tension: a level of accuracy

no human would ever notice is still insufficient for microsecond-scale radio synchronization by several hours' margin.

Time Error (TE) is the difference between a node's output time and true reference time, typically reported as a worst-case value

over an observation window. TDD networks require tight TE control because neighboring base stations share the same frequency

channel and time-multiplex uplink and downlink transmissions; misaligned frame boundaries beyond the network's error budget cause

uplink and downlink windows to overlap between cells, producing interference. ITU-T G.8271 formalizes this into accuracy classes,

with ±1.5 µs covering standard LTE-TDD/NR-TDD deployments and ±1 µs and below applying to some configurations and more advanced

coordinated radio features. This is an end-to-end budget shared across the full delivery chain from the primary reference to the

radio unit, so a node's own holdover error draws directly on the same margin consumed by every other stage of the chain.

• Frequency standard: https://en.wikipedia.org/wiki/Frequency_standard
• OCXO: https://en.wikipedia.org/wiki/Crystal_oven
• Rubidium standard: https://en.wikipedia.org/wiki/Rubidium_standard
• Caesium standard: https://en.wikipedia.org/wiki/Caesium_standard
• Hydrogen maser: https://en.wikipedia.org/wiki/Hydrogen_maser

• Adaptive drift compensation for holdover oscillators:

https://www.sitime.com/api/gated/Adaptive-Drift-Compensation-for-Holdover-Oscillators.pdf

Net Insight (Nasdaq: NETI B) provides the highest performing, most open video transport and media cloud technology for content

providers as the industry standard for flexibility and service across live contribution, distribution and remote production media

workflows.

For over 25 years, the world’s leading content owners, broadcasters, production companies, service providers and enterprises have

trusted Net Insight’s Emmy® Award winning Nimbra technology to guarantee media delivery. Today, Net Insight partners with hundreds

of customers in over 70 countries to ensure media flows across managed and unmanaged IP networks, and the cloud – from anywhere,

to everywhere. It enables customers to get the best from any mix of virtualized, cloud and IP technology and is the only platform

to support all the major industry standards, protocols and clouds.