Crosstalk Measurement Techniques for Multi-channel and Multi-rate High-speed Serial Communication Systems

Due to some drawbacks of parallel systems at high operating frequencies (such as skew, timing budget and layout constraints), many systems have turned to serial interfaces to transmit information. These serial interfaces can be designed to support multiple standards (such as SD-SDI and HD-SDI in digital video broadcast systems, USB and Firewire in data transmission systems, video streams at different frame resolutions/rates in dual-channel HDMI/DVI systems) and multiple data rates. In fact, different serial interfaces can transmit different standards simultaneously on multiple channels and be integrated on the same device (such as a four-independent channel SERDES). In this way, there will be high-speed signals switching at different rates on this device. This raises the question: "Is there interference between these signals?"

The interference caused by the high-speed switching of adjacent signals on the signal itself is called crosstalk . This effect manifests itself as jitter, which is the offset of the signal edge from its original position. Large amounts of this jitter can cause timing budget inaccuracies in parallel systems and can cause the clock and data recovery PLL to incorrectly recover data in serial systems.

Because crosstalk is harmful, it is important to calculate the amount of crosstalk that can occur in the worst case. Currently, there are no standard crosstalk measurement techniques in the serial world. This article will describe effective measurement techniques and how to determine whether the amount of crosstalk is acceptable for reliable data transmission.

Crosstalk

Crosstalk is the effect of adjacent signal traces on another signal.

The cross-coupling effect caused by the trace is called the victim. The trace that is measured is called the victim. The trace that cross-couples to the victim trace is called the aggressor. Figure 1 shows this relationship.

Crosstalk is related to the edge rate of the aggressor signal; faster edge rates result in more crosstalk. When the operating frequency of the transmitter increases, the transmitter usually increases the edge rate to improve the signal noise margin. Therefore, to measure the worst case, the aggressor channel needs to be switched to the highest frequency (see the sidebar: Advanced crosstalk measurements section: Aggressor frequency sweep).

Jitter

Crosstalk must be considered because it can be a major contributor to the amount of jitter in a device. Simply put, jitter is the deviation of a signal edge from its original position. Large amounts of jitter in a serial communication link can result in bit errors in the received serial bit stream.

Phase-Locked Loops

In 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, each with its own reference clock. One thing to note is that when adjacent PLLs switch between different frequencies, additional crosstalk may be introduced. This section briefly describes the structure of the transmit PLL and the tests that are performed on its performance. The receive PLL has a similar frequency response characteristic and is therefore tested similarly.

The transmit PLL is a clock multiplier (CMU). It takes in an input clock (REFCLK) and outputs a bit clock at ten times the frequency of REFCLK. Figure 2 shows a block diagram of a transmitter PLL. The amount of jitter that passes through the PLL depends on where the jitter enters the PLL. If the jitter enters through the REFCLK input, only the low-frequency components are transmitted through because the PLL acts as a low-pass filter . Low-frequency jitter has little effect on receiver performance because the clock and data recovery (CDR) PLL can track the low-frequency jitter and correctly receive the data. However, if jitter is injected in the middle of the loop (such as crosstalk between PLLs), the system acts as a high-pass filter. In this way, crosstalk between PLLs can become a source of unwanted, high-frequency jitter. To test the performance between the PLLs, three frequency variations between the victim and aggressor REFCLKs are performed: high frequency offset (> 100 MHZ), low frequency offset (< 1 MHZ), and nearly identical frequency (< 1 kHz) between the sources.

Jitter and Crosstalk Measurement Techniques

Crosstalk can be measured in two ways: jitter in the time domain and crosstalk in the frequency domain.

Measuring crosstalk in the time domain

To measure jitter in the time domain, the eye diagram of the victim channel can be observed on a high-bandwidth oscilloscope. An eye diagram is formed by superimposing waveforms whose phase relationship is determined by the phase difference of the trigger signal. The amount of jitter is obtained by analyzing the histogram formed at the visible crossing (Figure 3), which is the intersection of a positive edge and a negative edge. When a waveform crosses the histogram window, an "encounter" is recorded in the histogram. The histogram formed therein conforms to a Gaussian distribution.

Two important values ​​obtained in this measurement are the peak/peak jitter and the root mean square (RMS) value. Peak/peak jitter is the difference between the minimum and maximum times that the histogram meets. Because of the random nature of jitter, this value is uncertain, but it can provide useful information if measured over a long period of time. The RMS value (or standard deviation) converges quickly to a stable value. In a Gaussian distribution, it can be shown that the probability of a peak/peak jitter exceeding fourteen times the RMS value is less than once in a 1012 bit period. Therefore, to guarantee a 10-12 bit error rate, the jitter tolerance of the receiver should be greater than fourteen times the RMS value.

When measuring crosstalk in the time domain, the data pattern can be any pattern (as opposed to the specified periodic sequence mentioned in the frequency domain section). In this way, the jitter budget of the system in a real environment can be measured.

To determine the effect of crosstalk in the time domain, it is only necessary to measure the jitter increase of the victim channel without and with the aggressor.

Measuring Crosstalk in the Frequency Domain

To measure crosstalk in the frequency domain, a high bandwidth spectrum analyzer can be used. The amplitude of the reference frequency of the victim can be compared with the amplitude of the reference frequency of the aggressor. The reference frequency of a signal is the lowest intrinsic frequency of the signal, also known as the first harmonic. For example, the reference frequency of a 20MHz clock perfect square wave is 20MHz. In turn, it is composed of the sum of sinusoids composed of the products (or harmonics) of multiple reference frequencies. The most common square harmonics are odd harmonics. The first harmonic (20MHz in this case), the third harmonic (60MHz) and the fifth harmonic (100MHz) form the most common form of square waves. Therefore, if the crosstalk component is much smaller than the fifth harmonic, its impact can be basically ignored. [page]

When looking at the frequency response of a basic random data pattern, the spectrum will not only have peaks at the odd harmonics. Instead, the energy will be more distributed throughout the frequency range. In this case, it will be difficult to see the effects of crosstalk because the spectrum energy floor may be higher than the crosstalk energy. Thus, when measuring crosstalk on a serial data transmitter, the most effective pattern is a 1010-type periodic pattern. The entire pattern looks like a square wave, so the energy floor is lower when it is not at the odd harmonics.

Crosstalk Measurement Equipment Setup

Figure 4 shows an example of equipment setup for crosstalk measurement of a multi-channel high-speed serial transmission device, in this case a Cypress CYV15G0404DXB independent channel serializer/deserializer. The reference code clock (REFCLKx) for each channel is provided by a different Agilent 8133A pulse generator. RMS jitter of the pulse generator

should be low, since its jitter is directly related to the jitter of the serial output. Therefore, reducing the jitter of the reference clock makes it easier to see the effects of crosstalk on the serial data path. The Agilent 8133A has an RMS jitter of less than 5 ps (1 ps is typical). Measurements at the transmitter end are made on the serial output OUTA+. Other channels can be operated at independent data rates.

Time domain jitter measurements can be made using the Agilent 86100A high-bandwidth oscilloscope. Frequency domain measurements are made using the Agilent E4407B spectrum analyzer.

To measure the amount of worst-case crosstalk, all channels must be turned on and the transmitted signal must be fed back to the receiver to maximize the amount of crosstalk. When looking for the cause of crosstalk, you can turn on one channel at a time and observe the configuration that causes the largest jitter increase, or the configuration that causes the highest energy peak related to the frequency of the aggressor channel.

Conclusion

Using time domain measurements, we can better understand how crosstalk affects system performance. Since jitter is the cause of bit errors, this measurement helps determine the jitter budget of the system. Again, the data pattern can be any common data pattern (e.g., PRBS 23), so that real systems can be analyzed.

Frequency domain measurements can be a useful tool in determining the cause of crosstalk. The spectrum analyzer screen graph provides an easy way to detect non-original signal peaks. The frequencies of these peaks can be used to determine which aggressor signals are most affecting the system and determine where the crosstalk is occurring (PLL, signal trace, I/O buffer, etc.).

---

Sidebar: Advanced Crosstalk Measurements Section: Aggressor Frequency Sweep

This section will demonstrate that the effects of crosstalk increase as the aggressor edge rate increases. As described in the crosstalk section, transmitting devices typically increase their edge rates as the operating frequency increases. Therefore, the effects of crosstalk increase as the operating frequency increases.

The test equipment configuration is the same as the initial crosstalk measurement configuration, except that the spectrum analyzer is used to measure the serial output. For all tests, the code clock of the victim channel is operated at a fixed frequency of 150MHz. In contrast, the aggressor channel is swept and swung within the frequency range supported by the device (19.5MHz to 150MHz). The spectrum analyzer has a maximum retention function that retains the highest recorded energy at each frequency point. All channels transmit in a "1010101010" pattern.

Figure a shows the spectrum of the victim channel without an intruder. The reference frequency of the serial data is 750MHz. Figure b shows the spectrum of the victim channel when the code clock of the intruder channel is swept and swung over the full operating frequency range. The effect of crosstalk can be seen in the low frequency range (<150MHz). Although these values ​​are less than 1/100th of the reference frequency, the increase in amplitude can still be clearly seen when the frequency ranges from 20MHz to 150MHz.