SIGNAL INTEGRITY: Take care with programmable pre-emphasis

BY GEOFF LAWDAY -- EDN Europe, 01 Jun 2009

The capability to adjust pre-emphasis levels in serial-data streams is becoming universal; but you need to be careful, or it can do more harm than good.

Last time, in this column, I discussed the need to maintain fast edge rates, especially in the new LVDS (low-voltage differential-signal) serial buses. Going back to basics and applying Fourier analysis, I recalled that a fast edge is composed of a series of high-frequency sinusoidal-signal components. Consequently, any loss of the high-frequency signal components will degrade a signal edge, making a system susceptible to metastability. A good example of metastability is the set-up and hold violation data is clocked into a D-type latch: if the data or clock edge is degraded, the output of the latch can easily float between logic levels, which results in an indeterminate system state.

PCB (printed-circuit-board) dielectric losses are a principal cause of the attenuation of high-frequency-signal components, where the losses are due to both a capacitive and skin effects. Capacitive dielectric loss becomes the dominant factor in attenuation as frequencies rise because its effect is proportional to frequency, whereas losses due to skin effect are proportional to the square root of frequency. Both dielectric loss and skin effect can cause problems of inter-symbol interference because the attenuation effectively prevents the signal from reaching its full amplitude within the required duration, or its symbol time. As a result, the signal symbol—such as logic 1 or logic 0—will spread into the following symbol, mixing the symbols. The effect is pattern-dependent; engineers know it as PDJ (pattern-dependent jitter) or DDJ (data-dependent jitter). A further complication comes from dispersion, where the losses in the medium cause the different frequency components of the signal to arrive at the receiver out of phase and symbols to merge.

The simple solution would be to increase the signal strength to compensate for the attenuation. Unfortunately, increasing the signal strength does not solve the problem of the selective loss of the high frequencies. Also, the PDJ would deteriorate as each symbol would be unable to achieve its full strength within its allotted time slot and may spread even further into the next symbol—or slot—as a result of the increased signal level. Increasing the signal strength will also proportionally increase the noise and crosstalk in the system. What is more, the overall power consumption of the logic driver—or transceiver—will also rise, as the driver buffer will increase the amount of current flowing into the PCB trace.

A solution to the problem of highfrequency loss is to provide a method of boosting purely the high-frequency components of the signal, while leaving the low-frequency components in their original state. One way to achieve this is to use pre-emphasis. Pre-emphasis operates by boosting the high-frequency energy every time there is a transition in the data, because the data edges contain the highfrequency content of the signal and it is the signal edges that deteriorate with the loss of the high-frequencysignal components. Figure 1 shows a PCI-Express test signal with added pre-emphasis. Careful design of the pre-emphasis circuit can also help to overcome dispersion and reduce PDJ. Manufacturers who incorporate preemphasis filters within their devices are generally careful to design filters with a frequency cut-off that prevents each signal symbol from spreading into the next symbol space.

There are many different methods of defining and setting pre-emphasis. For example, the programmable preemphasis circuitry within a modern device typically has five different levels of pre-emphasis. It’s not possible to build an exact value of pre-emphasis into each device as the percentage preemphasis level is dependent on the individual PCB’s high-frequency losses, and too much pre-emphasis will increase system noise and cross-talk. Therefore it is important to measure and set the correct amount of pre-emphasis, or you could do more harm than good.

In my next article I will consider equalisation, which—in general—is a much better solution to the PCB highfrequency- loss problem.

Author Information
Dr Geoff Lawday is Tektronix Professor in Measurement at Buckinghamshire New University, in the UK. He is the coauthor of A Signal Integrity Engineer’s Companion (www.edn-europe/article. asp?articleid=2421).

 

Our Sponsors



Ads by Google