Measuring And Controlling Jitter In Digital Video Transmission

Transcript

Measuring and Controlling Jitter in Digital Video Transmission Systems A Tiernan White Paper May 3, 2001 Abstract This paper describes the use of simple video test equipment to measure those video parameters that are most affected by transmission jitter or “ cell-delay variation” in an MPEG-2 transport stream. Such jitter is encountered in poorly designed MPEG-2 encoders and decoders, but is more commonly associated with packet multiplexing systems, such as MPEG-2 transport stream multiplexers, ATM routers or IP-based routers. This paper describes how to use the results of those measurements to:  Ensure proper performance of the MPEG-2 encoders and decoders deployed in a particular application  Optimize the Quality of Service (QoS) parameters required for an ATM transmission system delivering MPEG2 video service By employing these techniques the broadcaster will be able to:  Ensure all encoders and decoders owned by the broadcaster are operating within specification  Measure the quality of service of the network, as leased by the broadcaster from a carrier. 2002 Radyne ComStream Inc. Radyne ComStream Inc. 6340 Sequence Drive San Diego, CA 92121 Phone: 858.458.1800 Fax: 858.657.5400 Website: www.radn.com PN 505-011985-0001A “ wander” , rather than jitter. This paper addresses both jitter and wander measurements. Packet & Cell Jitter in Compressed Video Transmission Systems These variations may be due to these different causes: The video timing of a typical MPEG-2 compressed video transmission system is shown in the following diagram. Video Timing Generator 27 MHz PLL Oscillator Composite Video Out 2. PCR generation errors within the MPEG-2 video encoder 3. Transport Stream packet propagation variations within the transmission system 4. Marginal STC recovery performance within the MPEG-2 video decoder 27 MHz Clock Recovery PLL 42 Bit STC Counter PCR Generator PCR Extract MPEG-2 Transport Stream Network Interface Timing variations in the video input signal (such as from VTR head wobble) MPEG-2 Decoder MPEG-2 Encoder Video In 1. MPEG-2 Transport Stream Network Network Interface MPEG-2 compressed video transmission puts the compressed video data into Transport Stream packets of 188 bytes each (184 bytes of payload and 4 bytes of packet header). Included in this Transport Stream are clock synchronizing parameters sent at regular intervals, around 10 to 60 times per second. These clock synchronizing fields, called Program Clock Reference (PCR), are the instantaneous value of a 42 bit long counter which counts the 27 MHz System Time Clock (STC) located in the MPEG-2 video encoder. This 27 MHz clock is locked to the incoming video signal. The PCR in the Transport Stream permits the MPEG-2 decoder to recreate the encoder’s System Time Clock. This recreated clock guarantees that the decoded video output operates at the same rate as the video signal input to the MPEG-2 encoder. If the time of arrival of the PCRs to the decoder varies with reference to the time the PCR was created from the encoder’s STC (ignoring the constant minimum propagation time from encoder to decoder), the recreated STC in the decoder will suffer variations. These variations cause the decoded analog video output to experience timing variations, which can be measured as line to line jitter, frame to frame jitter or phase variations in the recreated chroma subcarrier. In the decoder’s serial digital video output (SDI), these variations cause either timing jitter to the SMPTE 259M or SMPTE 292M serial bit timing, which is measured as bit jitter or wander. It is important to note that timing variations that vary at a rate lower than 10 Hz are usually referred to as 1 Most modern MPEG-2 encoders and decoders should perform very well when there is no propagation variation experienced within the transmission system. However, it is always prudent to verify proper performance before making a major capital equipment purchase. Methods for measuring encoder and decoder PCR performance are described in the next section. Many MPEG-2 compressed video transmission systems use Transport Stream multiplexers to combine several video programs into a single Transport Stream. These Transport Stream multiplexers may be separate units or they may be incorporated into the MPEG-2 video encoder itself. An MPEG-2 Transport Stream multiplexer can induce PCR jitter unless special logic functions have been incorporated to update the PCR values to correct for variations in delay as the PCR packet propagates through the multiplexer. Such "PCR correction” logic should be a part of any MPEG-2 Transport Stream multiplexer the user may wish to purchase. Before buying any Transport Stream multiplexing equipment, the customer should verify that it does indeed provide such a PCR correction function. The PCR jitter measurement techniques described in this paper can be used to verify the proper operation of any Transport Stream multiplexer’s PCR correction function. Since most PCR jitter problems are associated with variations in the propagation time of the transmission system, the majority of this paper addresses how to characterize and minimize these effects. In particular, recognize that SVC operation may provide different cell delay variation performance from one call to the next, depending on the nature of any other traffic that may be sharing each of the individual connections between each of the routers in the connection path. ATM cell routing has a very well defined effect on PCR jitter. The effects of ATM cell multiplexing at each routing point in the transmission system are analyzed and methods for minimizing these effects are presented. Telecommunication networks normally make a distinction between a network edge connection and a network internal connection. In North America, a network edge connection is usually at DS0 (64 kbps), DS1 (1.544 Mbps), DS3 (44.760 Mbps) or OC-3 (155.52 Mbps) bit rates. Professional quality video transmission systems normally use either DS3 or OC1 3 data rates for the network edge connection. Network internal connections for these professional video transmission systems normally use DS3, OC-3, OC-12 (640 Mbps), OC-48 (2.5 Gbps) data rates. Very soon OC-192 (10 Gbps) data links will become common. OC-768 (40 Gbps) data links are in development. Compressed Video Transmission Over Cell/Packet Based Networks A typical ATM or IP based network is shown below. Network Router Internal Connections Network Router Network Interface Network Router Network Network Router Edge Connection Network Interface Edge Connection In an ATM network, the information is delivered using ATM cells that are 53 bytes long, 48 bytes of payload and 5 bytes of header. The individual routers in an ATM network store the individual incoming ATM cells in a FIFO buffer. There is a separate FIFO buffer for each Virtual Circuit (VC) being managed by the router. If there are several Virtual Circuits using a single connection between routers, then the router multiplexes the cells onto the output connection according to a scheduling algorithm determined at the time each Virtual Circuit is set up. In ATM based networks, there are two types of Virtual Circuits:  Permanent Virtual Circuits (PVC)  Switched Virtual Circuits (SVC) A short analysis of the two common edge connection data rates is useful to illustrate the effect of ATM cell multiplexing on PCR jitter. DS3 Data Rate: At 44.736 Mbps, a 53 byte ATM cell requires: 53 bytes × 8 bits/byte / 44.736 Mbps ≅ 9.5 microseconds Therefore, if a particular ATM cell carrying the PCR information for a particular MPEG-2 Transport Stream arrives in the input FIFO buffer of an ATM router just one clock time too late, then the ATM cell must wait for 9.5 microseconds before it has its first chance to propagate to the router’s output. If the router’s scheduling algorithm decides that some other Virtual Circuit has higher priority, then the PCR cell will have to wait another 9.473 microseconds. Permanent Virtual Circuits are configured manually by the network provider and are always available to the user, 24 hours a day, seven days a week. They provide guaranteed connections that are always available, but they are more expensive. Switched Virtual Circuits are like telephone calls. The Virtual Circuit is established when the user makes the phone call, referred to as “ call setup” . When the user is finished, the Virtual Circuit is relinquished to the network provider via a “ call termination” . SVC operation is much less expensive, but the user must expect that they may not get the connection they request if any point in the route is congested due to other traffic. They must also 1 Some customers use multiple DS1 circuits grouped together by use of an “ inverse multiplexer” . If the customer only needs 2 to 20 Mbps of network capacity, then multiple DS1 circuits delivered to the customer premises is much less expensive than a single DS3 circuit. 2 Depending upon the following items, this ATM cell may have to wait several 9.5 microsecond cell intervals before it is finally allowed to proceed to the output:  Number of different Virtual Circuits open on the router’s output,  Configuration of each Virtual Circuit’s QoS parameters and  Nature of the scheduling algorithm employed within the router’s multiplexing function, A cell interval of 9.5 microseconds is almost 10,000 times the 1 nsec. jitter specification in SMPTE 170M. (See “ Rate of rotation” section below.) It represents a little over one seventh of a horizontal line time or almost 34 cycles of the chroma sub-carrier. This amount of jitter is too much for most MPEG-2 video decoders to manage. OC-3 Data Rate At 155.52 Mbps, a 53 byte ATM cell requires: 53 bytes × 8 bits/byte / 155.52 Mbps ≅ 2.7 microseconds Therefore, an ATM cell carrying the PCR information for a particular MPEG-2 Transport Stream may have to wait for 2.7 microseconds before it reaches the router’s output. Again depending on the configuration, several 2.7 microsecond intervals may pass before the cell is allowed to continue to the output. While this value is better than the interval for a DS3 rate multiplexer, it is still far too large for most MPEG-2 video decoders to manage. From these results it is apparent that faster data links suffer less PCR jitter. However, these same faster data links normally also carry more services on different Virtual Circuits, so each ATM cell must compete with more traffic for any particular output cell time in the multiplexer. This increases the likelihood that a particular cell may have to wait several cell times before it gets a turn. If the network transmission path passes through several different routers, then each router can introduce even more jitter. higher grade “ Quality of Service” (QoS) from the network provider. One method available to the network provider to provide better QoS cell delay variation performance is to allocate more bandwidth to a particular Virtual Circuit. This gives the scheduler more cell times to that Virtual Circuit to schedule a particular cell’s passage to the output. This approach helps, but at the expense of bandwidth, which costs more money. If the cell delay variation is over-specified, the user’s phone bill may double or even quadruple to pay for the additional bandwidth. In other instances, the additional bandwidth just isn’t available at all. Network “ edge products” are available, designed to connect between the network and the video decoder, which “ de-jitter” the received data stream. These devices normally work by either monitoring the fullness of a very large FIFO and using that information to control the output clock rate to the video decoder or by measuring the cell delay against an internal reference. Both approaches work well, but can add additional delay. Additionally, some methods of measuring cell delay may not be universally compatible with all ATM streams, or all Transport Streams. The Tiernan TATM8 MPEG-2 to ATM interface product uses to the FIFO fullness technique in order to ensure compatibility with any ATM stream. Measuring Video Timing Jitter Video transmission systems are characterized by the impairment of the video quality imposed on the video image at the receive end compared to the transmit end. The impairment across an analog video transmission system is characterized with the normal analog video performance measurements used in any video production or television studio. Digital video transmission systems impose impairments on the video image that may not be adequately characterized by these traditional analog video measurements. These systems require measurements specifically oriented towards the requirements of the digital video transmission environment. Sometimes traditional analog measurements can be substituted to infer performance issues associated with a digital video transmission link. The most straightforward method for measuring the network jitter as it affects the MPEG-2 Transport Stream is to use two Transport Stream analyzers, one on each end of the network, as shown below. If the packet/cell jitter is deemed to be excessive, the user should ask the network provider to improve the jitter performance of the network connection, either through an upgrade to the equipment in the network or by renegotiating a service level agreement with a 3 Composite Video In A/D Converter D/A Converter SMPTE 259M SDI Video In Composite Video Out SMPTE 259M SDI Video Out Video Compressor MPEG-2 Transport Stream Network Interface Video Decompressor Transport Stream Analyzer MPEG-2 Transport Stream Network Transport Stream Analyzer Network Interface Most Transport Stream analyzers available today will measure the PCR errors directly in nanoseconds, usually as a histogram. By comparing the PCR jitter at the output of the network with the PCR jitter measured at the input of the network, this error can be used as a direct measure of the maximum cell delay variation experienced by the Transport Stream as it propagates through the network. This jitter imposed by the network itself is used to verify the actual network ATM QoS parameter “ cell delay variation” as specified in the service level agreement between the network provider and the user. The problem with this measurement method is that Transport Stream analyzers are very expensive. Purchasing two Transport Stream analyzers can break the capital equipment budget for many broadcasters. Assuming the measured statistics of the PCR jitter at the input to the network do not vary significantly (usually a reasonable assumption), a single Transport Stream analyzer can be used to measure the PCR jitter at the input of the network. Then the analyzer can be transported to the output of the network to measure the received PCR jitter. This reduces the burden on the user’s capital equipment budget, but even a single Transport Stream analyzer can be very expensive, especially if it is only used to monitor the PCR jitter from the output of the network. The histogram can also provide some insight into the nature of the causes of the network jitter by analyzing where the peaks occur in the histogram. For instance, if they are at 2.7 microsecond intervals, then the jitter probably comes from an OC-3 connection. Similarly if the peaks occur at 9.5 microsecond intervals, then in probably comes from a DS3 connection. Peaks at other intervals indicate problems at points in the system operating at other corresponding bit rates. Indirect Measurement Using a Vector-scope If the user has access to an analog composite version of the input video signal and to the decoded output in 4 one location, such as using a network loop-back of the Transport Stream at the receive end of the network, this clock recovery error is easily measured using a simple video chrominance vector-scope. Connect the analog input video source into the “ reference” input of the vector-scope and into the input of the video transmission link equipment. Connect the output of the video transmission link, through a D/A converter, to the normal input of the vector-scope. Finally, configure the vector-scope to use the “ External Reference” input for the timing reference. The decoder’s clock recovery performance can be monitored by watching the vector-scope display’s rate of rotation. This setup is shown in the diagram below. Included is a Transport Stream switch that permits the user to measure the clock recovery performance of either the entire system or just the encoder and decoder without the network element. It is important to note that this approach induces twice the network PCR jitter because of the network loop-back. Due to the nature of the vector chrominance measurement, this technique is probably one of the most sensitive methods for measuring video jitter and wander. Composite Video In A/D Converter Video Compressor Network Loopback Ref In Vector scope Network Interface ATM Network Network Interface Video In Composite Video Loopback Video Decompressor D/A Converter When the vector display is referenced to the External Reference Input the vector waveform appears to rotate around the center of the display. This rotation is normal. It will spin very rapidly when the decoder first locks to the Transport Stream and then slowly settle down as the clock recovery function in the decoder locks to the incoming PCR values. This settling time will normally be several seconds, or even as much as five minutes. First, switch out the network and measure the clock recovery performance of the MPEG-2 encoder and decoder alone. Once the clock recovery function has settled down, the display will slowly rotate a few degrees per second. This settled rate of rotation is proportional to the variation in the arrival time of the PCRs with respect to the PCR value itself, along with the quality of the decoder’s clock recovery function. A longer integration time in the decoder’s clock recorder/player with excessive head wobble. If the same problem occurs when using a high quality test pattern generator or the output of the studio blackburst generator, then the problem is most likely in the MPEG-2 encoder or decoder. The best way to tell whether it’s the encoder or decoder is to measure the PCR jitter of the encoder with a Transport Stream analyzer. An alternative is to substitute a different brand of MPEG-2 video encoder or decoder and run the test again. recovery PLL low-pass filter will provide a slower rate of rotation. Some decoders use a variable integration time low-pass filter, which uses a fast integrator during PCR acquisition and a longer integration time for PCR tracking. This approach provides both fast PCR acquisition and good filter performance in the presence of PCR jitter. All Tiernan decoders use a multi-stage approach to clock recovery for this reason. Rate of rotation: The NTSC specification, ANSI/SMPTE 170M (successor to the old EIA RS170 specification) section 11.1, specifies the chroma sub-carrier drift must be less than 0.1 Hz/sec. and the peak to peak jitter must be less than 1 nsec. over one horizontal line. Translating these values into degrees of rotation on the vector scope gives: Once the performance of the rest of the system is characterized, the ATM network with the loop-back at the remote end is switched back into the test. The additional jitter caused by the network is measured in degrees peak-to-peak. The low frequency wander is measured in degrees of rotation per second. 0.1 Hz/sec. × 360 degrees/Hz × 2 = 72 degrees/sec. The MPEG-2 encoder/decoder jitter is subtracted and the resulting network jitter is converted from degrees peak-to-peak into nanoseconds peak-to-peak using the inverse of the jitter formula above: and 1 nsec./horiz. line × fsc × 360 × 2 = 0.000000001 × (5,000,000 × 63/88) × 360 × 2 ≅ Network_Jitternsec = (Total_Jitterdegrees - MPEG_Jitterdegrees) × 1000 / (5 × 63/88 × 360 × 2) 2.6° per horiz. line. Since this is a loop-back test the network jitter will be twice as large as the one way path. Therefore an extra factor of two is included in the above formula. One disadvantage to this method is that it assumes the jitter statistics are similar for the return path as for the forward path, which may or may not be accurate. Note that SMPTE 170M is a studio specification and the values specified in it are appropriate for studio quality NTSC video where all components are synchronized to each other, normally by means of a master “ black-burst” signal generator. Therefore, it may be unrealistic to expect the decoded output of every MPEG-2 video transmission system to always meet the letter of this specification. However, as a customer, one should ask the equipment and network suppliers to use these values as reference points and to approach them as best as possible. A network with jitter beyond these limits by even a factor of two is not particularly serious. Nevertheless, these measurements will allow the user to rate the relative video performance of equipment from various suppliers. This formula assumes that the peak-to-peak statistics add arithmetically. The average jitter statistics would add geometrically, in a root-mean-square fashion. Assuming the clock recovery filter in the MPEG-2 decoder is a proper low-pass function, the actual PCR jitter of the incoming Transport Stream must be at least this amount. If this result exceeds the ATM QoS parameter “ cell delay variation” , as described in the network service level agreement, then the network is not performing according to the required specification. Because of integration effects of the indirect measurement method, if the measured jitter even approaches the limit specified in the service level agreement, then the actual cell delay variation may be exceeding the specified value. Some automated video measurement systems provide the ability to measure the long-term sub-carrier stability. These measurements may be used to substitute for the vector-scope. They also have the advantage of not requiring the external reference input. This is because the video measurement system regenerates it’s own reference sub-carrier derived from the input. This allows it to be used to measure the jitter of a one-way circuit, where a loop-back function is not available. The disadvantage is that an automated video measurement system is much more expensive than a vector-scope. Also note that if the performance does not appear to be adequate for the MPEG-2 encoder and decoder combination even without the network, this may be an indication that the video source itself may have a time-base problem. This can occur in a video tape 5 user’s manual, a telephone call to the manufacturer may be required. Indirect Measurement Using Waveform Analysis If the user does not have access to both the input and output of the video transmission link in one location, either a delayed sweep oscilloscope or an automated video measurement system can be used to measure the clock recovery accuracy at the output of the video decoder. The delayed sweep oscilloscope, preferably with a video trigger function, can measure the horizontal timing jitter. The automated video test system can measure either the chroma burst phase or the horizontal sync timing jitter. Both are an indirect measure of the decoder’s clock recovery accuracy. Composite Video In A/D Converter D/A Converter SMPTE 259M SDI Video In SMPTE 259M SDI Video Out Video Compressor Network Interface Video Decompressor D/A Converter Composite Video Out Several automated video measurement systems provide the ability to measure line-to-line jitter and even field-to-field or frame-to-frame jitter. These measurements may be used in the same manner as the oscilloscope measurements. Digital Video Output Bit Jitter: SMPTE 259M and SMPTE 292M both specify the maximum permitted bit clock jitter over a frequency range related to the fundamental clock frequency for each interface. SMPTE RP 184 [6], SMPTE RP 192 [7] and SMPTE EG 33 [8] define serial data jitter and how to measure it in a serial data interface. Oscilloscope SDI Analyzer Video Decompressor Network Longer measurement integration times can be achieved by delaying multiple horizontal lines or even entire fields or frames. However, the longer the delay, the greater the strain on the jitter & wander requirements of the oscilloscope’s time-base function. Network Interface SDI Analyzer Oscilloscope The video decoder and D/A converter located on the transmit side of the network are used for quantifying the amount of jitter from sources other than the network itself. The peak-to-peak jitter measured on the transmit side of the network is subtracted from the peak-to-peak jitter measured on the receive side of the network, so as to isolate the jitter component contributed by the network itself. The oscilloscope is configured to trigger on the horizontal sync tip, with the time-base delay function set to approximately 63 microseconds, until the next horizontal line sync tip edge is visible. The delayed sweep is set somewhere in the range of 10 nsec/div to 10 microsec/div, depending on the amount of observed jitter on the sync tip edge. This measurement is easier to observe with a digital storage oscilloscope, especially one with a long persistence capability. Ideally, the persistence should be on the order of two to four PCR intervals in the Transport Stream. Furthermore, the measurement requires a high quality time-base function in the oscilloscope. This time-base must have a low jitter and wander specification or else it will only measure the jitter and wander associated with the oscilloscope rather than from the video transmission system. Time-base jitter and wander specifications may not be easy to obtain for any particular make or model of oscilloscope. If this information is not included in the 6 Serial digital video measurement instruments from several manufacturers include indicators if the serial digital video source under test has excessive jitter. These instruments can be used to measure the PCR jitter of the MPEG-2 video decoder. Again, because of the integration effects of the clock recovery PLL in the decoder, the jitter at the SDI video output of the decoder will be much smaller than the PCR jitter on the Transport Stream input. Therefore, if the measured jitter at the SDI video output is even close to the network limit, the actual network cell jitter performance may be out of compliance. At least one of the serial digital video performance test instruments presently available indicates a compliance error if it measures excessive timing wander below 10 Hz. However, since no standard presently exists for serial digital video timing wander, this compliance error alarm is not necessarily significant. Rather, it should be regarded merely as an indication of potential problems in the video transmission system, not as an explicit compliance error. Conclusion As the video production and broadcast industry makes the transition from analog to digital video, the facilities will contain a mixture of both analog and digital video equipment. In order to verify video performance in a mixed analog and digital video transmission link, the user should account for the following issues: Bibliography 1. “ MPEG-2: Systems” , ISO/IEC 13818-1: • Cell delay variation is a very important parameter in ATM based compressed video transmission systems. • Cell delay variation is more of a problem for networks with longer cell transmission intervals. Therefore, higher data rate networks will normally provide better video jitter performance. • Direct measurement of network jitter requires expensive test equipment. • Many of the performance issues associated with an ATM based video transmission link can be characterized, or at least indirectly inferred, using existing analog video test equipment. This helps minimize the capital investment required for digital video test equipment, at least through the short term. • Generic Coding of Moving Pictures and Associated Audio Information: Systems, 7/95 (also ITU-T Rec. H.222). 2. ANSI/SMPTE 170M-1994, Composite Analog Video Signal – NTSC for Studio Applications, October 19, 1994. 3. ANSI/SMPTE 259M, 10-Bit 4:2:2 Component and 4fsc Composite Digital Signals – Serial Digital Interface, September 25, 1997. 4. ANSI/SMPTE 292M, Bit-Serial Digital Interface for High-definition Television Systems, May 7 1996. 5. SMPTE RP 184, Specification of Jitter in BitSerial Digital Systems, December 1, 1996. 6. SMPTE RP 192, Jitter Measurement Procedures Packet or cell based routing networks which use ATM cell routers can introduce excessive timing jitter into the decoded video output. In this case, the user should negotiate with the network provider for either an upgrade to the network equipment or for a higher grade of QoS. An alternative is to use network edge devices that incorporate network timing jitter reduction methods. in Bit-Serial Digital Systems, December 1, 1996. 7. SMPTE EG 33-1998, Jitter Characteristics and Measurements, February 1, 1998. Furthermore, it is important to recognize that ATM based networks will normally provide better timing jitter performance than IP based networks because the ATM cell is much shorter than typical IP packets. Jitter performance over IP based networks should improve if the streaming protocol uses shorter packets, at the expense of additional packet overhead. 7 Radyne ComStream Inc. 6340 Sequence Drive San Diego, CA 92121 Phone: 858.458.1800 Fax: 858.657.5400 Website: www.radn.com 2002 Radyne ComStream Inc