Advertisement

Article

Measuring crosstalk in high-speed serial links

Due to the pitfalls facing parallel systems at high operating frequencies (skew, timing budget, and layout constraints), many systems are converting to a serial interface to transport information. These serial interfaces may be designed to support multiple standards, such as SD-SDI and HD-SDI in digital video broadcast systems, USB and Firewire in data transfer systems, video streams with different frame resolutions/rates in dual HDMI/DVI systems) and thus, multiple data rates.

In fact, there are serial interfaces that can simultaneously transmit different standards across multiple channels, integrated into one device, (e.g. a quad, Independent-channel SERDES). Therefore, the device will have high-speed signals switching at different rates. This situation raises the question: “Will there be any interference between these signals?”

Crosstalk is the effect on a signal caused by the high-speed switching of a nearby signal. This effect can manifest itself as jitter, which is the deviation of a signal's edge from its expected location. A large amount of jitter can cause a timing budget failure in a parallel system, or it can cause a clock and data recovery PLL to incorrectly recover the data in a serial system. Due to the negative effects of crosstalk, it is important to determine the amount of crosstalk that exists during the worst-case scenarios. Currently, there are no standard crosstalk measurement techniques for the serial domain.

This article describes effective measurement techniques and how to determine if the amount of crosstalk is acceptable for reliable data transfer. The techniques include measuring the device's jitter output with a wide-bandwidth oscilloscope and spectral output with a high-bandwidth spectrum analyzer. It also describes configurations which yield the highest crosstalk scenarios, along with actual measurement data of a multi-channel, independent rate device. The measurements are performed at video serial digital interface (SDI) data rates because of the common application of transmitting multiple video standards, but apply to any high-speed standard.

Crosstalk and jitter
Crosstalk is the effect on a signal trace caused by cross-coupling with a neighboring trace. The trace that is being measured is referred to as the victim. The trace that is cross-coupled with the victim trace is referred to as the aggressor. A picture of this relationship is shown in Figure 1 .



Figure 1: Victim and aggressor traces, relative position on PCB

(Click to enlarge image)

Crosstalk is dependent upon the edge rate of the aggressor signal; a faster edge induces more crosstalk. When the operating frequency of a transmitter is increased, the transmitter usually increases its edge rate to improve the signal's noise margin. Therefore, to test the worst-case situation, the aggressor channels should switch at the highest frequency (See the “Advanced Crosstalk Measurement Section” below, for more detailed ananlysis.)

Crosstalk is a concern because it can be a major contributor to the amount of jitter present in a device. Simply stated, jitter is the deviation of an edge from its expected location. A large amount of jitter in a serial communications link may cause bit errors in the received serial bit stream.

Phase-Locked Loops
Within any serializer/deserializer (SERDES) device, there is a transmit PLL and a receive PLL. A quad, independent-channel SERDES has four transmit and receive pairs, where each pair has its own reference clock. A concern is that when adjacent PLLs are switching at different frequencies, additional crosstalk may occur. This section will briefly describe the structure of the transmit PLL and the tests that will be performed to measure its performance. The receive PLL has similar frequency response characteristics and will be tested similarly.

The transmit PLL, Figure 2 , is a clock multiplier unit (CMU).



Figure 2: Transmit PLL block diagram

(Click to enlarge image)

It receives an input clock (REFCLK) and outputs a bit clock that has ten times the frequency of REFCLK. The amount of jitter that propagates through a PLL depends on where the jitter entered the PLL. If it enters via the REFCLK input, only low-frequency components will pass through, because the PLL acts as a low-pass filter. Low-frequency jitter has a smaller impact on receiver performance because a clock and data recovery (CDR) PLL will track the low-frequency jitter and correctly receive the data.

However, if jitter is injected inside the loop (by crosstalk between PLLs, for example), the system acts like a high-pass filter. Therefore, crosstalk between the PLLs can be a cause of harmful, high-frequency jitter. To test the performance of the PLLs, three frequency variations between victim and aggressor REFCLKs will be performed: large frequency offsets (>100 MHz), small frequency offsets (<1 MHz), and nearly identical frequencies (<1 kHz) from different sources.

Jitter and Crosstalk Measurement Techniques
Measuring crosstalk can be performed in two ways, using jitter in the time domain and crosstalk in the frequency domain. To measure jitter in the time domain, the eye diagram of the victim channel's output is observed on a wide-bandwidth oscilloscope. An eye diagram is formed by a superposition of waveforms. The phase of these waveforms with respect to one another is determined by the phase difference with the trigger signal. The amount of jitter will be obtained by analyzing the histogram formed at the eye crossing (Figure 3 ).



Figure 3: Jitter measurements at the eye crossing

(Click to enlarge image)

The eye crossing is the point at which the positive edge and negative edge intersect. When the waveform intersects the histogram window, a “hit” is recorded in the histogram. The histogram in the figure resembles a Gaussian distribution.

The two important values obtained from this measurement are the peak-to-peak jitter and the root mean square (rms) value. The peak-to-peak jitter is the difference between the minimum and maximum time of hits in the histogram. This value is non-deterministic because of the nature of random jitter, but can provide helpful information if the test is performed for a long period of time.

The rms value (or standard deviation) converges to a stable value quickly. Assuming a Gaussian distribution, the peak-to-peak jitter will exceed fourteen times the rms value less than one time in 1012 bit periods. Therefore, to ensure a 10-12 bit error rate, the jitter tolerance of the receiver should be at least fourteen times the rms value.

When testing for crosstalk in the time domain, the data pattern can be any pattern, as opposed to specific periodic sequences in the frequency domain section. This way, the true jitter budget in a real-world system can be tested.

To determine the effect of crosstalk in the time domain, measure the increase in jitter on the victim channel between no aggressors and with aggressors, Figure 4 .



Figure 4a

(Click to enlarge image)



Figure 4b

Time-domain comparison between (a) aggressors disabled (a) and (b) aggressors enabled


(Click to enlarge image)

In this example, the victim channel is transmitting a pseudorandom built-in self-test (BIST) sequence at 1485 Mbps. The aggressor channel (when enabled) is transmitting the highest-possible frequency signal, which for this device is a 1010 pattern at 1500 Mbps. The rms jitter (also called the standard deviation) increased from ∼13.5 ps with no aggressors to ∼14.2 ps with aggressors. This translates to an increase in total jitter of ∼10 ps when targeting a BER of 10-12 .

Measuring Crosstalk in the Frequency Domain
To measure crosstalk in the frequency domain, use a wide-bandwidth spectrum analyzer. The amplitude of the victim's fundamental frequency (first harmonic) is compared with the amplitude of the aggressor's fundamental frequency. The first, third, and fifth harmonics are responsible for most of the shape of the square wave. Thus, if the crosstalk components are significantly smaller than the fifth harmonic, their impact will be negligible.

When looking at the frequency response of a fairly random data pattern, the spectrum will not only have peaks at the odd harmonics. Instead, the energy will be distributed across the frequency range. In this case, it will be difficult to see the effects of crosstalk because the spectral energy floor may be higher than the crosstalk energy. Therefore, when measuring crosstalk on serial data transmitters, the most effective pattern is a 1010 periodic pattern. This pattern will look like a square wave and thus, will have a low energy floor when not at the odd harmonics.

Figure 5 shows the spectrum of a victim channel that is transmitting a 1111100000 pattern at 270 Mbps (SD-SDI serial signalling rate) while the aggressor channel is transmitting a 1010101010 pattern at 1485 Mbps (HD-SDI serial signalling rate).



Figure 5: Spectral plot of victim channel

(Click to enlarge image)

In this case, the victim channel is equivalently transmitting a 27 MHz square wave (270 Mbps/10 bits per cycle) and the aggressor is equivalently transmitting a 742.5 MHz square wave (1485 Mbps/2 bits per cycle). Figure 4a shows the energy (in dB) vs. frequency from 0 to 800 MHz; every gridline on the energy axis represents 7dB.

Since this a real-world signal, there are some even harmonics, but they all have <-50dB (0.3%) of the energy of the fundamental. Figure 4b zooms in on the 125-165 MHz region to show the largest crosstalk peak in the victim channel. This peak is at 148.5 MHz, which is the character rate of the aggressor channel.

The other peak is the 5th harmonic of the victim signal at 135 MHz (5 times 27 MHz). In this case, the effect of crosstalk is nearly negligible because the largest crosstalk peak has ∼-28dB (4%) as much energy as the 5th harmonic. In Figure 4a, the aggressor's serial data fundamental frequency (742.5 MHz) can also be seen. It has approximately the same energy as the peak at 148.5 MHz. All other crosstalk peaks above 1 GHz are smaller.

In the above example, it shows the effect of crosstalk from the aggressor's character clock and from the aggressor's serial data. Crosstalk from the aggressor's character clock may be because the character clock traces are routed to close together. Crosstalk from the aggressor's serial data may be because the serial I/Os are not isolated well enough or that the serial PCB traces are too close together. If there was crosstalk at the aggressor's bit clock frequency (in this case, 1485 MHz), then the PLLs may be interfering with each other. This information can help the board or chip designer narrow down the cause of the crosstalk.

Crosstalk Measurement Equipment Setup
Figure 6 shows an example of a crosstalk measurement equipment setup for a multi-channel, high-speed serial transmission device, which in this case is a Cypress CYV15G0404DXB Independent Channel Serializer/Deserializer.



Figure 6: Crosstalk measurement equipment setup example

(Click to enlarge image)

The reference character clock for each channel (REFCLKx) will be supplied by a separate Agilent 8133A Pulse Generator. The rms jitter from the pulse generator should be kept low because its jitter directly relates to the jitter of the serial outputs. Therefore, minimizing the reference clock jitter will make it easier to see the effect of crosstalk on the serial data path. The Agilent 8133A's rms jitter is less than 5 ps (1 ps typical). The measurements performed on the transmit side will be taken at the serial output OUTA+. The other channels can operate at independent data rates.

The time-domain jitter measurement is performed with an Agilent 86100A Wide-Bandwidth Oscilloscope. The frequency-domain measurement is performed with an Agilent E4407B Spectrum Analyzer. The screenshot in Figure 3 is from the oscilloscope and the screenshot in Figure 4 is from the spectrum analyzer.

To determine the worst-case crosstalk, all channels should be enabled and the transmitted signal should be looped back into the receiver to maximize the amout of crosstalk. When debugging the cause of the crosstalk, the best approach may be to enable one channel at a time and observe the configuration that causes the largest increase in jitter or causes the highest energy peak at a frequency related to the aggressor channel.

Advanced Crosstalk Measurement Section: Frequency Sweep of Aggressors
This section will prove that the effect of crosstalk increases as the aggressor's edge rate increases. As described in the crosstalk section, transmitting devices usually increase their edge rate as the operating frequency increases. Thus, the effect of crosstalk should increase with operating frequency.

The test equipment setup is the same as the first crosstalk measurement setup, except only the spectrum analyzer is used to measure the serial output. For all of the tests, the victim channel's character clock will operate at a constant frequency of 150 MHz. Conversely, the aggressor channels will sweep across the complete operating range (19.5 MHz to 150 MHz) of the device. The spectrum analyzer has a Max Hold function that maintains the highest recorded energy for each frequency point. All channels will transmit a “1010101010” pattern.

Figure 7 shows the spectral plot of the victim channel with no aggressors.



Figure 7: Victim traces

(Click to enlarge image)

The fundamental frequency of the serial data is at 750 MHz. Figure 8 shows the spectral plot of the victim channel while the aggressor channel's character clock sweeps across the whole operating frequency range.



Figure 8: Aggressor traces

(Click to enlarge image)

The effect of crosstalk can be seen at the lower frequency range (<150 MHz). While these values are less than 1/100th the size of the fundamental frequency, it is clear that the amplitude increases with frequency from 20 to 150 MHz.

Conclusion
Using the time-domain measurement methodology, we can get a better idea of how crosstalk is impacting the performance of the system. Since jitter is the cause of bit errors, this measurement will help to determine the jitter budget of the system. Also, the data pattern can be any normal data pattern (e.g. PRBS 23), which enables real-world system analysis.

The frequency-domain measurement technique is a useful tool for determining the cause of the crosstalk. The spectrum analyzer provides an easy way to detect peaks that are not part of the original signal. The frequency of those peaks can be used to determine which aggressor signals are causing the largest impact on the system and to determine where (PLL, signal traces, I/O buffers, etc.) the crosstalk is occurring.

References
1. “HOTLink Jitter Characteristics,” Cypress Semiconductor application note, March 11, 1999. http://www.cypress.com/ cfuploads/support/app_notes/jitter.pdf
2. “Measuring Crosstalk in LVDS Systems,” Elliot Cole. Texas Instruments, January 2000, http://focus.ti.com/lit/an/ slla064/slla064.pdf

About the Authors
Jeff Hushley is a Senior Applications Engineer at Cypress Semiconductor Corporation, San Jose, California. He has 3 years of experience working with high-speed serial digital communication devices in broadcast and consumer video applications. He has a Bachelors in Computer Engineering from the University of Toronto. He can be reached at fre@cypress.com.

Palani Subbiah is a Senior Staff Applications Engineer at Cypress Semiconductor Corporation, San Jose, California. He manages an Applications Engineering organization that focusses on High-Speed Video Interconnect products. His expertise includes Signal Integrity, High Speed Serial Measurment techniques, Serial Protocols and Digital Hardware Design. He has Bachelors in Electronics from Sri Venkateswara College of Engineering, India and Masters in Electrical Engineering from Univerisity of Missouri ” Rolla. He can be reached at pds@cypress.com.

1 comment on “Measuring crosstalk in high-speed serial links

  1. oidjfvoijf
    July 29, 2015

    Are you a real estate agent looking for marketing ideas in 2015?

    My primary piece of advice is that you need to put your customers' needs first. While all forms of marketing should encourage your audience to eventually do business with you, your approach needs to be subtle and less self-serving. In 2015, your marketing strategy should focus on humanizing your business and automating things when you can.

    I recently put together an infographic that outlines how agents spend their marketing dollars and compared it to where they're getting their leads from. As I researched this topic, two things became very apparent:

    75 percent of a real estate agent's business comes referrals and word of mouth.
    But agents are still spending way too much money on new lead sources, rather than investing in their existing clients.
    Simply put, if most of your business is coming from your past clients and it costs you six to seven times more to acquire a new customer than it does to retain an existing one, then you should focus your efforts on marketing to your past clients.

    The post I wrote last year titled “7 Marketing Ideas for Real Estate Agents in 2014” quickly became the most viewed post on our site (over 20,000 unique viewers). So for 2015, I came up with 10 marketing ideas that agents can use to really help grow their businesses.

    1) Be Mobile and Responsive
    In 2014, I stressed that going mobile was important in terms of you posting to your social networks while on the go. This is still true today, but this year, I want to stress the importance of being mobile accessible.

    You cannot avoid it anymore. Your website and emails need to be designed for all devices and screen sizes and be responsive. Sixty-six percent of emails are opened on a phone or tablet. Facebook has more than 650 million daily active users on mobile devices. And mobile traffic to websites now accounts for almost 30 percent of visits. In short, there are plenty of reasons why you need to be sure that your business is mobile ready.

    How It Helps: Being mobile-friendly ensures whoever reads your emails or visits your website has a good experience and that this experience creates a positive perception of your business.

    2) Use GIFs in Email
    You've all seen GIFs before; they're those soundless graphics that automatically play on a loop. They're great for getting a quick laugh, but these days, they're more often used to caption a situation and convey human emotion without using words. For example…if it's Monday morning, and I'm having a tough time getting motivated, I might text or email this GIF to my friends:

    Marketing ideas – GIF use in email

    Did you know that you can use GIFs in email too? Our friends at Litmus have put together an awesome guide to using GIFs in emails. While you shouldn't use them in everything you send, they do add a comical element you just can't get from static images.

    Even though GIFs have been around since the late '80s, they've had a resurgence in popularity. GIFs are a fun way to help show off the human side of your business.

     

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.