Some smoothing of a noisy topic: suggestions for reducing channel-to-channel crosstalk.
AT LEAST FOUR significant problems contribute to eye collapse and an increased bit error rate in high-speed serial links. Our previous two columns addressed how reflections and losses affect ISI ISI International Sensitivity Index, see there , collapse of the eye, increased deterministic jitter and higher bit error rates. This month we look at how channel-to-channel crosstalk contributes to ISI and collapse of the eye.
Crosstalk from one channel to another can range from as high as 20% in some cases to as low as 0.01% in other cases. And just how much is too much, of course, depends on the received signal strength and how much noise margin is needed at the receiver. In an extreme case, the received signal may be as low as -30 dB from the transmitted signal and still acceptable. If a signal-to-noise ratio of at least 10 dB is desired, then the crosstalk should be less than -40 dB. This is less than 1%.
This is the basis of a simple rule of thumb. Unless you have an extreme channel, the channel-to-channel crosstalk should be kept below -40 dB. If it is ever larger than -40 dB, it may be important to perform analysis to see how much you can tolerate in your specific design.
One reason crosstalk discussions in the literature are so confusing is that so many design factors influence the magnitude of crosstalk. One could easily imagine a situation when it is a killer problem and other cases when it is trivial. It all depends on the specific design conditions and the criterion of how much is too much.
For example, consider a very bad, pathological situation, as illustrated in FIGURE 1. Two differential channels are routed in long lines in FR-4, but isolated from each other. However, there is a con-nector or package with 2" of trace in microstrip. Although it is designed as a 100 [ohm ohm (ōm) [for G. S. Ohm], unit of electrical resistance, defined as the resistance in a circuit in which a potential difference of one volt creates a current of one ampere; hence, 1 ohm equals 1 volt/ampere. ] differential impedance, all four signal lines are tightly coupled with a spacing equal to their line width. This occurs very often in flex connector strips.
[FIGURE 1 OMITTED]
With a rise time of 25 psec psec
picosecond , typical of a high-end PCIe gen II part, and a coupled length of 2", the FEXT (Far End Cross(X) Talk) A measurement of crosstalk between two wire pairs taken at the far end of the line. See ELFEXT and NEXT. would be about 20%. In this pathological configuration, the far-end noise is generated close to the aggressor TX, so the rise time is the shortest, and the FEXT noise travels to the victim RX very close to the connector, so it suffers little attenuation Loss of signal power in a transmission.
The reduction in level of a transmitted quantity as a function of a parameter, usually distance. It is applied mainly to acoustic or electromagnetic waves and is expressed as the ratio of power densities. .
The poor victim RX is at the end of a long, lossy See lossy compression.
(algorithm) lossy - A term describing a data compression algorithm which actually reduces the amount of information in the data, rather than just the number of bits used to represent that information. interconnect, so the received signal is attenuated. If the two channels are asynchronous Refers to events that are not synchronized, or coordinated, in time. The following are considered asynchronous operations. The interval between transmitting A and B is not the same as between B and C. The ability to initiate a transmission at either end. , the noise on the victim line is uncorrelated with the received bit, so the noise will spread uniformly over the entire unit interval.
This 20% FEXT noise, spread over the unit interval, can completely close a marginally acceptable eye, as in Figure 1. In some cases, channel-to-channel crosstalk can ruin your day, and it can arise in board traces, via stacks, connectors, packages and cables.
What can you do to reduce crosstalk to an acceptable level? TABLE 1 contains some suggestions. But, remember, you have to do your own analysis.
TABLE 1. Suggestions for Reducing Crosstalk DESIGN FEATURE DIRECTION FOR BETTER OR WORSE Aggressor TX and victim RX Asynchronous generally worse synchronous or asynchronous Channel-to-channel spacing Larger is better FEXT: microstrip or stripline Stripline is better NEXT: microstrip or stripline Stripline is better for channel to channel spacing >2x line width Differential impedance Lower impedance is generally better Interleaved or non-interleaved Microstrip: best is interleaved Stripline: best is non-interleaved Line to line coupling Tighter is always better. With an adjacent return plane, however, coupling is a second or third order factor; not critical Coupling length Shorter is always better Rise time Longer is always better Bit rate Usually gets worse, as rise time decreases
DR. ERIC BOGATIN is a signal integrity evangelist with Bogatin Enterprises (beTheSignal.com), a LeCroy Co.; eric@beTheSignal.com.