Measuring Wander in Video Distribution Systems

Tom Tucker, Tektronix Television Test

Introduction

The steadily decreasing cost of the technology required to transport video signals through non-traditional terrestrial transmission has created a virtual explosion in the deployment of affordable, high quality video distribution networks. As more is learned about the behavior of video signals traveling such networks, the need to monitor video sync timing is fast becoming a concern of network service providers. Specifically, the measurement and control of video wander, (defined as synchronizing signal phase variations below 10 Hz) within a video distribution network is necessary to ensure the quality of the video signal recovered at the output of the network.

Standard definitions and measurement methodology for baseband video wander is only recently being developed because video wander has never been a problem for "studio-quality" video processing. This is because studio-quality video processing almost always employs timebase correctors to lock video signals to a stable house reference.

However, more and more there are applications where video arrives at the studio as a serial-digital or analog composite signal directly from a codec where it is desirable to record this signal without a time-base corrector or frame synchronizer. In this case, problems with picture shifts and composite color hue errors can occur due to network induced wander in the synchronizing pulse and color burst timing. This can even be a problem with component serial digital applications since when the video signal is converted to composite the wander is not easily removed.

This paper examines sync timing wander and it's effect on the video signal. Additionally, this paper will describe a proposed specification for video wander, based on existing specifications/recommendations on horizontal sync jitter and subcarrier drift-rate, and a method of measurement based on a well known baseband video measurement instrument.

Digital Synchronous Network Overview

For some digital synchronous network operators the idea of providing video transport services is not a new topic. In fact, in many countries they have worked closely with broadcasters to provide dedicated links to transport television signals in analogue form. The advent of digital video provided many benefits including, the multiplexing of video with other information (eg. voice and data) over the same network. Since the same network backbone can be used for all activities, dedicated links are not needed. Additional advantages are also gained from using standardized interfaces, integral network management and easier maintenance.

Within these networks contribution quality video codecs are used at the high clock rates necessary for handling large volumes of video data. A large variety of video signals and formats can be transported, including analogue composite and digital component video.

SONET networks, which are synchronized to an external timing source, buffer their data and use pointers to indicate the start of the next part of the payload. At each step where data is multiplexed, the network may add or drop a few bytes of buffered data to maintain the average payload bit-rate. However, pointer adjustments cause a phase variation in the SONET network. A portion of this phase variation, i.e. wander, is usually transferred to the video signal that is recovered at the output of the codec in use. If this happens, a composite video signal can exhibit subtle shifts in color hue or variations in sync phase when viewed on a waveform/vector monitor. Excessive pointer adjustments in the network can produce levels of video wander to accumulate causing more severe problems including loss of synchronizing lock-up.

Irrespective of the type of synchronous network used to transport the video signal, for example SDH or SONET, the unique method of maintaining timing of these networks can induce unwanted timing impairments into the video delivery service. Therefore, demands on network timing and synchronization performance will be at their highest when providing video delivery services.

Timing and Synchronization Performance Required By Video

In order to understand the impact of the digital synchronous transport network timing wander on video service quality, the timing requirements of the video service first need to be fully understood.

Beginning with the video signal specification, Tektronix has derived a new jitter/wander template by analysis of the colour subcarrier specifications. Any video network should deliver a signal conforming within these limits if studio timing quality is to be preserved. RP-154-1994 from SMPTE specifies a black-burst, studio timing reference signal. It currently has no burst jitter spec but does specify H-sync jitter at ±2.5 ns. We suggest an equivalent subcarrier-burst jitter limit to be ±0.25 ns.

PAL System I colour subcarrier drift-rate is specified at ±0.1 Hz/sec or 0.0226 ppm/sec. This is not a wander spec but implicitly specifies a 12 dB/octave sinusoidal wander limit to remain within the drift-rate.

Colour subcarrier tolerance is specified as ±1 Hz or ±0.226 ppm for PAL (±10 Hz for NTSC). Again, this implicitly specifies a 12 dB/octave sinusoidal wander limit to remain within the frequency tolerance.

Derivation of a Measurement

The measurement and specification of video wander is often simplified or ill-defined such that consistent measurements are nearly impossible. For example, what does it mean to specify H-sync jitter as 10 nsec max?? What is the reference? If you measure the leading of sync with a triggered oscilloscope triggered on the displayed pulse, it will always be less than 10 nsec no matter what the real jitter. If you delay out 100 lines then you are measuring the delay time base jitter and the signal jitter relative to a point earlier in time. The effect of this is to highpass or differentiate the low-frequency components of the jitter. Jitter spectral components with periods of 200-lines are measured with a gain of two and spectral components higher than that are aliased. Clearly, this is not a preferred method.

What if you use a free-running reference to trigger you scope? In this case the jitter, or phase-error, is changing constantly at a fixed rate due to the frequency-error and jitter can't be easily separated. If you use a PLL to trigger your scope, then you are removing the low-frequency components of the jitter.

In the PLL case, what should the PLL bandwidth be for a meaningful measure? If it is too high, then the measurement will not include large low-frequency components that may cause problems for certain equipment. In real systems the magnitude of the low-frequency jitter or wander is often very large and the lower you go for the PLL bandwidth the larger the jitter measurement will be possibly exceeding the 10 nsec level by a factor of 1000 or more. Since virtually every system always re-locks in some way to the video, this large, very low-frequency jitter or wander component is usually not important. However, it could be a problem with a composite tape recorder which may record a hue error while trying to track this wander. It is not clear how low the PLL bandwidth should be to effect a meaningful measurement.

At the other extreme, is it meaningful to talk about jitter spectral components above half the line-rate (15,734/2)? Clearly, there is no meaning to the jitter spectrum above 1/2 the rate of the signal being measured other than the fact that high-frequency signals that could modulate the H-sync phase will be mixed down to the band from DC to Fh/2.

There is a least one specification on H-sync jitter. SMPTE RP-154 1994 specifies that sync jitter for a studio black-burst signal should be less than 2.5 ns as compared to the past average of at least one field (262 samples). This processing creates a second-order, high-pass filter with a 50 Hz corner frequency. Therefore, jitter much larger than 2.5 ns is allowed below that frequency. If you average for 1000 fields, then most sources would fail since it would restrict wander at 0.05 Hz rates to less than 2.5 ns. How many fields should you average to make a useful measurement?

The previous discussion raises some important questions since there currently is no wander definition, specification or measurement method for video. However, one can be derived from existing specifications and recommendations on H-sync jitter and the color subcarrier drift-rate. For example, SMPTE 170M recommends that the color subcarrier frequency not drift any faster than 0.1 Hz/sec (0.028 ppm/sec). Also, in the context of studio quality video, there is an approximate static phase relationship between the sync leading edge and color-burst, referred to as SCH phase, precluding any significant difference between the subcarrier and H-sync phase wander. This then puts the drift-rate limit of 0.028 ppm/sec on the leading edge of H-sync as well. For sinusoidal jitter x(t) = Asin(2 pi ft), this drift-rate limit can be viewed in terms of a sinusoidal spectrum accommodation plot of A versus f, common in video jitter analysis, as follows:

Note that the peak jitter can increase as 1/f² and still be below the drift-rate limit. For example, the video sync could be phase-modulated with a 0.01 Hz sinewave with a peak phase deviation of > 1 µs and not exceed the drift-rate limit. However, in order to comply with the drift-rate value, jitter/wander spectral components above 1 Hz must have a peak value less than 1 ns and decrease in peak amplitude as 1/f². This is an absurd requirement since jitter below 1 ns peak is not discernible even if it has fast transients that cause the frequency drift-rate to be far in excess of the limit. Therefore, in the spectral region above 1 Hz, we must look for another requirement with a more practical limit on peak jitter.

SMPTE RP-154-1994 is a recommended practice limiting the amount of peak jitter on the leading edge of H-sync for an analog black-burst video reference signal to 2.5 ns. More importantly, it defines how this value is determined by comparing the time x of the 50% point of the leading edge of sync to it's average x_ave over at least one TV-field. Let H(f) = y/x be the sinusoidal gain of the process y = x - x_ave. Then limiting y < 2.5 ns limits x < 2.5/H(f), where x = Asin(2 pi ft).

Notice that H(f) is, when the average is delay-compensated, effectively a second-order, high-pass filter on the jitter, allowing large amounts of jitter below the corner frequency. Since RP-154 leaves open how many fields to average (determines the corner frequency), it is not clear how many should be averaged for a standardized measurement methodology. However, since jitter below 2.5 ns peak is considered good enough it is reasonable to consider using RP-154 with at least a 90-field average for a jitter measurement above 0.5 Hz and the drift-rate limit for a wander measurement below 0.5 Hz. Therefore, for baseband video, a reasonable demarcation frequency to separate sync jitter from sync wander or drift-rate is 0.5 Hz. Consider the following diagram:

It is also interesting to note that the RP-154 jitter accommodation below the corner frequency is the same slope as the drift-rate requirement and, in particular, when the average is 90-fields, it overlays the drift-rate curve with a 2.5 ns floor above about 0.5 Hz. Therefore, by using RP-154 with a 90-field average, we have a specification and measurement method that is simultaneously measuring SMPTE-170M drift-rate conformance below 0.5 Hz and RP-154 peak jitter above 0.5 Hz. To better understand this, consider the typical video signal with broad-band jitter and wander that is composed of many spectral components. Let it be measured according to RP-154 with a 90-field average where the maximum peak output is detected over a period of 1000 seconds. Those wander or low-frequency jitter components below 0.5 Hz down to about 1 mHz contribute to the output such that the detected value will exceed 2.5 ns if they exceed the drift-rate limit. Simultaneously, any jitter components above 0.5 Hz with peak values above 2.5 ns will also cause the detected output to read above the 2.5 ns limit. A problem with this method is that it effectively sums jitter and wander spectral components into one output value leaving some small uncertainty as to which is causing it to exceed the limit when a signal is equally dominated by both. In fact, maybe both jitter and wander are just below their respective limit and they just happen to be additive.

Because of this, it is probably preferable to separate the wander (drift-rate) and jitter measurements into two outputs allowing each to be more or less independently quantified. Clearly, this can be done by measuring the video sync leading edge jitter with a 0.5 Hz bandwidth high-pass filter of order 3 or more. The wander can then be separately measured by processing the sync leading edge data with a second-order differentiator below 0.5 Hz and at least a single pole roll-off above 0.5 Hz.

The process derived above defines a measurement methodology by defining the spectral bands and performance limits for the measurements of;

1. Jitter,
2. Drift or Frequency-error, and
3. Drift-rate.

These measurements are diagrammed below:

Jitter, Drift (frequency-error), and Drift-rate Measurement for H-sync

Note 1: Third-order, 0.5 Hz bandwidth, low-pass filter to prevent aliasing and separate wander measurements from jitter effects.

Note 2: Decimation of line-rate phase samples to frame-rate to remove video correlated phase errors from drift and drift-rate data.

Note 3: Frequency error or drift with 1 PPM per µsec scale factor.

Note 4: Drift-rate of frequency with 1 PPM/sec per µsec scale factor.

Note 5: Third-order HPF to separate jitter data from frequency drift and drift-rate data.

The previous diagram describes the signal processing steps to implement three measurements on timing errors of the leading edge of the horizontal sync of a television signal. These three measurements are defined as follows:

1. Jitter. The spectral components of the total jitter above 0.5 Hz where the typically large components below 0.5 Hz are substantially rejected so as not to contribute to this measurement. This measure is consistent with RP-154 when using a 90-field average, but provides better rejection below 0.5 Hz by using a third-order high-pass.
   
   
2. Drift-limit (or frequency-error). This measurement is the first derivative of the jitter below 0.5 Hz and is the instantaneous frequency error due to jitter components below 0.5 Hz. When the reference frequency is well known, this output replaces a frequency counter for determining the frequency accuracy of H-sync. Jitter components correlated with frame and field are removed from the measurement as well as spectral components above 0.5 Hz (Jitter as measured in #1). Typically this measurement has a static result when viewed with the relatively coarse scale factor necessary to resolve the NTSC limit of ± 2.8 ppm.
   
   
3. Drift-rate. This measurement is the second derivative of the jitter below 0.5 Hz and is the first derivative or rate-of-change of the frequency-drift in #2. This is the most important measurement since high rates of frequency shift may cause PLL/servo tracking errors (hue-errors) in tape machines trying to remain locked to the video. Also, since the drift-rate spec on subcarrier is quite tight, it is often the parameter that is exceeded first with normal types of phase perturbations.

A Solution To The Industry

The three jitter/wander measurement methods described can be measured with the VM700T Video Measurement Set's Horizontal Jitter measurement application and the new option 22 Drift Rate measurement application.

1. Jitter. Use the current VM700T "JITTER" measurement. This measurement was originally designed for VCR transport testing. However, it does effect a high-pass filter in the region of 5 Hz and provides a display of H-sync jitter correlated with field and frame. H-sync jitter measurements made in this mode will generally agree with those made with the ideal HPF response proposed earlier unless the wander components are very large.
   
   
2. Drift. Use the new VM700T option 22 "Drift Rate" measurement. This new measurement conforms precisely with the proposed drift-limit measurement described in this paper. The limit for NTSC is 2.8 ppm or 2800 ppb and markers are included at both limits. The cursors indicate the peaks over the past 17 seconds and the values are indicated in ppm and Hz @ Fsc.
   
   
3. Drift-Rate. Use the new VM700T option 22 "Drift Rate" measurement. This is the primary measurement and conforms precisely with the Drift-rate measurement described in this paper. In this mode you can monitor the effects of both Drift and Drift-rate aspects of the H-sync wander. The drift-rate limits for NTSC (± 28ppb/sec) are clearly marked. As in the Drift measurement the cursors indicate the peaks over the past 17 sec and readout in ppm/sec and Hz/sec @ Fsc.

Conclusion

While it is clear that SONET and SDH networks have the capability to provide a common transport infrastructure for voice, video and data services well into the 21st century it is also clear that the timing accuracy of video transported over these networks requires special attention. Specifically, the instantaneous frequency error and, more importantly, the rate of frequency drift of the video horizontal timing must be held within acceptable limits to assure reliable utilization of the video signal within a professional video facility.

This paper defines the requirements of a baseband video Jitter and Wander measurement template, related to the transport of video signals over SONET and SDH digital networks, based on current professional studio timing specifications. Additionally, a simple yet effective method for measuring video horizontal sync timing Jitter and Wander, based on the popular VM700T Video Measurement Set has been developed which allows both network operator and their customers to ensure the quality of video transport services from end to end.

Navigate Press Materials