Pendulum telecom network synchronizationjitter measurementwander measurementTIE MTIE TDEV

Telecom Network Timing: Jitter, Wander and 1-PPS Verification

From raw TIE data to MTIE and TDEV, from verifying a GPSDO's 1-PPS offset to comparing eight sync sources at once: put numbers on network timing problems.

In short

In telecom networks synchronized by SyncE, PTP and GNSS, timing quality comes down to two quantities: jitter above 10 Hz and wander below 10 Hz. The Pendulum CNT-91R timestamps any clock below 160 MHz with 35 ps resolution to measure TIE, and TimeView 3 computes the ITU-T MTIE and TDEV from it; when verifying a GPSDO's 1-PPS, the same setup resolves a -2.9 ns offset with 0.6 ns random uncertainty. The CNT-104S has four channels on one common time axis with resolution below 7 ps, and two synchronized units compare 8 1-PPS signals at once. This article explains which problem each measurement solves and why the reference decides how far the result can be trusted.

Jitter and wander: the 10 Hz dividing line

The phase deviation of a network clock from an ideal clock is called jitter when it is fast and wander when it is slow, and the industry draws the line at a phase-variation frequency of 10 Hz. The sources differ: jitter is mostly cycle-to-cycle noise, while wander comes from slave clocks disciplined to an upstream signal, master clocks disciplined to GNSS, any frequency-control loop containing a PLL or FLL, and slow changes in propagation delay on long fiber runs caused by day-night temperature swings. A free-running, temperature-stable cesium, rubidium or OCXO oscillator does not generate wander by itself.

The consequences of excess wander are concrete: frame slips in SDH and SONET, retransmissions that cut throughput, dropped calls at handover in mobile networks, and pixel errors or frozen frames in video. IP networks were originally unsynchronized, but SyncE and PTP carry frequency and time through disciplined slave clocks, and the same wander problems come with them. The first step in managing these effects is to measure them as numbers.

TIE is the base quantity: CNT-91R with TimeView 3

Multi-channel TIE trend plot in TimeView — TIE trend plot

The base quantity of wander measurement is TIE, the time interval error between the edge of the clock under test and the edge of an ideal clock. The CNT-91 timestamps every edge under test with 35 ps resolution, the software derives the ideal clock from the reference timebase, and the difference is the TIE sample by sample. The method applies to any clock signal below 160 MHz, with no dedicated interface needed for 2.048 MHz or 1.544 MHz. TimeView 3 then computes the MTIE and TDEV defined by ITU-T, ANSI and ETSI from the raw TIE, plus ADEV and MADEV.

MTIE is the maximum peak-to-peak TIE within each observation interval; it only ever increases, and it is compared against the ITU-T masks to judge frame-slip risk. An MTIE rising linearly with observation time usually indicates a frequency offset in the clock under test. TDEV is the rms dispersion of TIE within each interval, reflecting short- to medium-term stability and periodic modulation, and it correlates with bit-error risk. The example in the document is a 12000-second record of a GNSS-disciplined rubidium oscillator: the slope corresponds to a frequency offset of 3.5×10⁻¹¹, and at the end of the observation interval MTIE reaches 430 ns, about 88% of a 2.048 MHz clock period.

The RTIE and MRTIE trap: the reference must not wander

A TIE curve that is a straight line with a slope means a fixed frequency offset between the clock under test and the measurement reference, but the curve alone cannot say which side is off. Removing that line gives RTIE, and the peak-to-peak of RTIE is MRTIE. Pendulum's document gives an example: the MTIE of one signal set reached 600 ns because of the frequency offset, while MRTIE with the offset removed was only 9 ns. The gap is large, but MRTIE cannot be compared against the ITU-T MTIE masks and cannot reveal a network clock drifting after losing lock.

Simple instruments with an inadequate reference often report MRTIE instead, hiding the offset of their own timebase. That is exactly what to avoid. The correct order is to make the measurement reference itself free of wander first, either with the CNT-91R's built-in rubidium timebase or a calibrated atomic reference connected externally to the CNT-91, then use MTIE as the primary parameter and treat MRTIE as supporting information for observing disciplining-loop behavior. The calibration uncertainty of the reference enters the systematic uncertainty of TIE directly as 2×10⁻¹¹ times the observation time, and no post-processing can remove it.

Verifying a GPSDO's 1-PPS: offset, stability and convergence time

Allan deviation versus averaging time — ADEV stability curve

A GPSDO's 1-PPS output should align with the top of the GPS second, and verifying it needs a reference 1-PPS at a known position plus a counter that resolves sub-nanosecond differences. Pendulum's evaluation paper measures the time interval between the rising edges of the reference and the 1-PPS under test with a CNT-91R: random uncertainty 50 ps rms, systematic uncertainty from channel mismatch below 500 ps, with TimeView capturing up to 250,000 samples per second and plotting time interval against time.

The measured run shows what matters in this kind of measurement. After a cold start the 1-PPS under test was initially offset by 338 ms; the position fix completed after 262 seconds and the offset dropped to around -2.8 ns; in steady state the mean was -2.9 ns with a standard deviation of 0.6 ns. The roughly 3 ns propagation delay of a 60 cm RG58 cable then has to be compensated, giving a true corrected offset of -5.9 ns. Over the whole observation the reference 1-PPS had an MTIE of only 280 ps and rms jitter of about 40 ps, against 2.3 ns for the unit under test, which confirms that the nanosecond-level steps in the time interval curve come from the device under test, not the reference.

Multi-channel 1-PPS phase comparison: the common time axis of the CNT-104S

A synchronization room often has several 1-PPS sources to compare with each other, and pairwise measurement is slow and never shows them all at the same instant. The CNT-104S has four parallel input channels whose four timestamp engines run on one time axis with time resolution below 7 ps, and each channel performs accumulated time interval or timestamp measurements directly on inputs A, B, D and E. With the sample interval set to 0, every 1-PPS edge is recorded: 86401 samples is one day, 604801 is one week, and a single channel stores up to 10 million samples, or 115 days of continuous recording.

To compare 8 signals, two CNT-104S units cannot simply sit side by side, because their time axes are unrelated. Pendulum's application note solves this with block arming: the pulse output of the first unit goes through a power splitter to the external arming inputs of both, both time axes zero at the arming instant, and every timestamp afterwards is relative to that common zero. The time-axis uncertainty between the two units is below 3 ns, typically 1 ns; arming through a measurement channel instead brings it down to typically 50 ps at the cost of reducing the devices under test to 6. Using an external UTC 1-PPS as the arming source references the whole timestamp set directly to UTC.

The modulation domain: plotting frequency against time

An oscilloscope shows voltage against time, a spectrum analyzer voltage against frequency, and modulation domain analysis frequency against time. Lock-in overshoot, synthesizer settling time and frequency modulation of an oscillator by supply ripple are either invisible on the first two instruments or visible only as a blurred edge. The CNT-91 measures continuously with zero-dead-time timestamps, 250,000 samples per second in free-running mode and up to 1.85 million samples in one capture, and TimeView plots them as frequency, period, phase, pulse width or duty cycle against time.

For fast, repetitive transients, TimeView's repetitive sampling shifts each measurement 10 ns further from the trigger point step by step, reaching an equivalent sample rate of 100 MSa/s; the UHF VCO example in the document measured a frequency swing of about 29 MHz and a rise time of 10.7 µs. An FFT of frequency-against-time data puts modulation frequency on the horizontal axis, and the output jitter of one pulse generator analyzed this way showed a clear 100 Hz component pointing at the power supply. Histograms have diagnostic value too: a Gaussian distribution is random jitter, a bathtub shape is sinusoidal modulation, and two bars are square-wave modulation.

The measurement reference and how Evencat can help

Every measurement above rests on one premise: the reference must be better than the device under test. The FTR-210R locks GNSS and a rubidium oscillator together, with 1 pps time accuracy to UTC better than 10 ns rms and standard outputs of 5 × 10 MHz, 1 × 5 MHz and 1 × 1 pps. It serves as the shared 10 MHz external reference for two CNT-104S units and as the UTC arming source for an eight-way comparison. During GNSS loss, holdover drift is typically about 1 µs per 24 hours, so a brief antenna blockage does not invalidate a week of records.

Evencat is the authorized Pendulum distributor in Taiwan and can help decide how to divide the work between the CNT-91R and CNT-104S for your network architecture, configure the reference and prepare calibration documents, and provide technical support for TimeView and SCPI automation; acceptance tests and short projects can also start with frequency counter rental.