EYE-OPENER+™ Self Equalized Conductor Technology
Comments on the Value of EYE-OPENER+TM Conductor Technology
1. The following points should be remembered when speaking about EYE-OPENER+TM conductor/cable:
a) The core/skin design of EYE-OPENER+TM conductor provides a distributed or per unit length equalization, without resistor – inductor - capacitor (RLC) circuitry required on the connector paddlecard.
b) Clearly, RLC equalization circuits can be designed to perform similarly to EYE-OPENER+TM conductor (even better, actually). This is assuming, however, that the discussion is limited to looking at a single data line.
c) In practice, multi-line, high-traffic signal configurations are the norm. EYE-OPENER+TM conductor becomes superior to RLC equalization in these situations, particularly when high-density connectors are used. What happens is that the RLC equalization circuits for each data line must be placed in close proximity in these types of connectors. This proximity creates serious crosstalk problems, since the inductors are generating fields that inevitably “talk” to each other.
d) When a cable vendor provides a spectrum of cable assembly lengths for a particular application, the longer assembly lengths are usually equalized. Equalization circuits must be tuned for a specific cable length, and it is critical that the appropriate equalization circuit is installed in the right cable. Otherwise, the result can be worse (in terms of SI quality) than if no equalization circuit were installed at all. EYE-OPENER+TM conductor technology eliminates these types of inventory/human error problems, since the distributed equalization is engineered directly into the cable itself.
2. Raw conductor is only one component of cable performance – the dielectric material also plays an important role. Gore’s ePTFE dielectric material has a dielectric constant of 1.3, while our competitors are using materials with a constant of approximately 1.6. At the longer cable lengths, this difference is measurable, as can be seen in the following plots generated by a third-party connector company:
EYE-OPENER+TM
No Equalization
RLC Equalization
Figure 1 - Eye Patterns: Equalized and non-Equalized
*Same connectors used in all three assemblies – EYE-OPENER+TM and standard assemblies used a non-equalized PCB, while the RLC equalization assembly used a PCB to accommodate this circuitry.
3. Note that the EYE-OPENER+TM assembly has approximately 25% more amplitude than the RLC Equalization assembly. This is due to the performance benefit gained from using a low-loss dielectric material, and the impedance “tuning” that can be achieved without sacrifices in the size of the cable.
4. EYE-OPENER+TM conductor technology enhances the performance of signal conditioned (i.e. pre-emphasis) systems. Jitter is well known to be the limiting factor for maximum signal transmission length in multi-gigabit systems. The whole idea behind equalization (either EYE-OPENER+TM or RLC circuit) is to reduce jitter and, thus, improve eye opening. Jitter is reduced in equalized cables because the transfer function (frequency response) of the cable is “flattened” in the bandwidth pertinent to the data rate that will be transmitted by the cable.
a) In pre-emphasized systems, the situation is more complex, since the effective transfer function is the superposition of the pre-emphasis transfer function and the cable loss transfer function:
Figure 2 - Transfer Function is a superposition of pre-emphasis and cable functions
b) In order to look further at EYE-OPENER+TM and pre-emphasis, it is necessary to understand more about how pre-emphasis works. Pre-emphasis is actually “de-emphasis”, in the sense that the latter part of the signal waveform is suppressed to exaggerate the leading edge of the square wave. This is, in essence, “signal equalization” done to compensate for the low-pass filter effect of the cable. EYE-OPENER+TM (and RLC equalization) work to solve this same problem from the hardware (cable) side.
Figure 3 - Standard driver pulse shape compared with pre-emphasis pulse shape
c) Clearly, this signal “de-emphasis” places a premium on loss, since the signal is deliberately reduced in amplitude with this approach. The key benefit to EYE-OPENER+TM conductor is that it lowers the “level” of pre-emphasis required to achieve clarity in the eye-pattern at the end of a given cable length. This is because the frequency flattening response of EYE-OPENER+TM works in concert with the signal-level “de-emphasis” to alleviate the low-pass filter characteristic of the signal transmission path as a whole. The benefit of this synergy is seen in the plot below. Note that EYE-OPENER+TM cable requires a lower level of pre-emphasis (50%) than standard conductor (120%), and achieves a lower-loss attenuation trace with a flatter response in the bandwidth of interest.
Figure 4 - S21 Transfer Function of pre-emphasis / cable system
d) It is important to note that conventional RLC equalization would provide a similar complement to the signal-level pre-emphasis conditioning, but not without problems. Again, the inductors present in these types of circuits generate fields that create connector NEXT – particularly with the introduction of even more voltage transitions by virtue of the signal-level pre-emphasis.
Viability of a Fixed-Level Pre-Emphasis Scheme: Standard Cable Conductor Vs. EYE-OPENER+TM Conductor
Purpose: This experiment was performed to evaluate the viability of a signal conditioning scheme which allows a single level of pre-emphasis for all lengths of cable. The performance of standard conductor cable vs. EYE-OPENER+TM conductor cable in such a scheme was also evaluated.
Equipment: Marvell IB 88X2040 IB demo board, WL Gore IBNTST412X test fixtures, Agilent 86100 oscilloscope, DC blocks, Gore AWG-24 IB4000-2,12,15 standard conductor cables, Gore AWG-24 IB4000-2,10,15 EYE-OPENER+TM cables.
Set-up: One end of the cable under test was connected through a Gore test fixture to the 86100 via DC blocks. The other end was connected through a Gore test fixture to Channel “0” of the Marvell demo board. The Marvell board was set to generate a CJTPAT data pattern with a data rate of 2.5Gb/s. The amplitude was set to the default level, and the pre-emphasis level was set to 120% for standard cables or 67% for EYE-OPENER+TM cables. The clock output of the demo board was connected to the 86100 mainframe trigger. The pk-pk jitter measurements were taken using the automatic jitter measurement of the 86100 where possible - the cursors were used when necessary. The eye height measurements were all done by manually placing the cursors on the eye diagram. All measurements were performed on Channel “3” of the cable assembly.
Results: The following table shows the results of the jitter and eye height measurements:
|
Length/Type |
2m |
5m |
12 |
15 |
2 |
10 |
15 |
|
Jitter |
96ps |
100ps |
76ps |
89ps |
66ps |
67ps |
56ps |
|
Eye Height |
360mV |
368mV |
300mV |
244mV |
537mV |
385mV |
296mV |
Figure 5– Jitter and Eye Height Data Table for Fixed-Level Pre-Emphasis Schemes (no signals running on adjacent lines)
Specifying a single pre-emphasis level, while enabling the link to work at longer lengths, has an associated cost in eye amplitude. At short lengths, the excessive pre-emphasis causes a diminished eye height. However, cable losses are small at the shorter lengths, so the eye height is still sufficient (see figure 6). Jitter, however, poses a greater problem in the single-level pre-emphasis scheme. At short lengths, there is a large amount of jitter caused by the “over pre-emphasis” at those lengths. In the data table above, it can be seen that the 2m and 5m lengths of standard cable are near the limit for jitter allowed by the InfiniBand specification. It is important to note that the jitter values in the table above were collected with no data signals running on adjacent lines. This means that the jitter values in the table are optimal – i.e. the jitter values would be somewhat higher in the presence of additional data traffic (connector NEXT would contribute some jitter). EYE-OPENER+TM conductor cable shows much less jitter across the length spectrum, because the pre-emphasis level required to achieve a given maximum cable length is lower. This is substantiated by the fact that, for this testing, the optimal pre-emphasis level for a 15-meter standard cable (based on eye height) was 120%, while the optimal level for the 15-meter EYE-OPENER+TM conductor cable was only 67%.
Figure 6- Eye Pattern of a 2-meter cable with standard conductor @ 120% pre-emphasis level
Discussion: The strategy of using a single level of pre-emphasis is viable from the standpoint of eye height. The cable losses at short lengths are small enough that even a highly “over pre-emphasized” signal will have sufficient eye height.
Jitter limits the level of allowable pre-emphasis, and this level may be less than the optimum level required for maximum cable length. It appears that, for standard cable, the ultimate constraint is the jitter level at the shortest cable length, rather than the eye amplitude desired at the maximum cable length.
The use of EYE-OPENER+TM conductor cable allows for lower pre-emphasis levels, and this removes the jitter constraint at the shortest cable lengths. The pre-emphasis level can simply be adjusted to achieve the maximum link length.