USB 3.0 is the next-generation version of the USB standard. Capable of data rates to 5 Gbits/s, it’s currently being developed by Intel and a sponsor group (www.usb.org/developers/docs). One concern with this emerging standard is the integrity of signals passing through (for example) a printed-circuit board (PCB), a long cable, and the associated connectors. To guarantee acceptable performance, you must include either fixed or adaptive equalization in the system. Adaptive equalizer circuits are complex and require a long training sequence. Fixed equalization is simpler, easier to design, and less power hungry.
Experimental verification shows that fixed (vs. adaptive) equalization can achieve simplicity, lower cost, and lower power consumption while enabling the transmission of 5-Gbit/s data streams over a path consisting of an FR4 PCB trace, USB 3.0 cable, and connectors. As shown in Figure 1, you can assemble a simple test setup using 10 inches of FR4 PCB material with copper traces, a transmit equalizer, a USB 3.0 cable, and a receive equalizer.
A bit-pattern generator sends an IKJPAT bit stream to the FR4 PCB. Similar to CJPAT, IKJPAT is an 8B/10B pattern designed to stress a timing-recovery circuit. Two ICs, though not designed for USB 3.0, can extend USB 3.0 transmissions beyond the three meters specified for that standard: You can implement the fixed transmit equalization with a transmit equalizer (MAX3982) that offers four pre-emphasis levels.
The receiver shown is a 1.8-V fixed passive equalizer (MAX3785) followed by a limiting amplifier. Both have been off-the-shelf products for the last few years. Similar devices are available from vendors such as National Semiconductor and Texas Instruments. The fixed receive equalizer is a balanced T network followed by a limiting amplifier (Fig. 2).1
When you run the test patterns, an unequalized signal fed through 10 inches of FR4 and three meters of USB 3.0 cable shows considerable degradation (Fig. 3). With fixed equalization, however, and even after increasing the cable length to six meters, the observed signal shows a very good eye pattern (Fig. 4a). The better eye pattern implies better signal quality, and thus fewer chances for misinterpreting the data (fewer bit errors).
If you zoom in on the equalized signal of (Fig. 4b), you can see that the peak-to-peak jitter is just 36.4 ps.
Shortening the cable to the 3-meter length used for the unequalized test, you can see that the fixed-equalization scheme produces an eye pattern that remains stable and wide open (Fig. 5a). When you zoom in on the crossover region (Fig. 5b), you still get a very clean transition. Thus, the cable interface seems to have little need for the more expensive adaptive-equalization circuitry.
Reference:
1. For complete details on a fixed-equalizer design, refer to “Designing a Simple, Small, Wide-band and Low-Power Equalizer,” Clark Foley, DesignCon, 2003.