Other Parts Discussed in Thread: ALP, DS110DF410, DS110DF410EVM, DS110DF410EVK
Hi Team,
Please tell me about the EOM.
My customer and I are displaying the eye pattern in the EXCEL sheet, by reading register 0x25/0x26.
When I build up the eye pattern, the eye is not in the center, a little bit shifted to the right.
How can you read the data which eye pattern is positioned in the center ?
Is there any setting ?
Best Regards,
Kawai
Hi Michael-san,
Please let me ask you one more about EYE position.
Our customer had read the register and built the EYE diagram such as below.
The center of the vertical position of the Eye is shifted down side.
Would their be an EYE diagram such as above, or is it a mistake in the reading procedure ?
Best Regards,
Kawai
Hi Kawai-san,
When the eye monitor is used, the retimer CDR acquires lock by locating a point in time where the average bit transitions occur. This point in time is assumed to be the equivalent of the 0.5UI point, and it is ideally where PDAC (variable-phase clock) = 0. (Note that there are 64 evenly spaced phases that make up one UI.) However, a perfectly located average bit transition location rarely occurs because there is a small delay between the input signal and where the CDR clock is sampling, resulting in the eye appearing offset. For example, it may be found that the average bit transition happens somewhere around PDAC = 20.
Having located the average bit transition point in time, the center of the eye is assumed to be roughly 30-31 phase divisions afterward. In an ideal eye monitor, this would mean that on Column 32, the actual 0.0UI point would be perfectly in the center. However, in the second example where PDAC = 20 at the average bit transition location, the actual 0.0UI point would be located around Column 50.
When the eye monitor is measured, the device initially assumes the perfect scenario where the eye center is located around Column 32 and that there are approximately 30-31 columns on each side of the "eye center" to fill with bit error data. However, in imperfect cases like the second example, if we started with Column 50 and assume that there are 30-31 columns to fill on the right, we will hit Column 63 and still have 16-17 columns left to fill. These extra columns that cannot fit on the end are then populated on the front few columns instead. If you look at the screenshot of PRBS-31 data running from a 10.3125Gbps source, you can see that this horizontal offset leads to extra columns beyond Column 63 being placed in the first few columns.
Having an internal eye monitor that is not centered is normal. As of now, ALP's algorithm appears to be locked in a .dll file, so it will take us some time to uncover exactly how ALP is recentering the diagram. However, it appears that if the retimer is conscience of which PDAC setting the average bit transition time occurred, then the eye monitor simply performs a horizontal shift so that the eye appears centered again.
The downward shift from your latest post seems strange. If the input signal is a perfect differential signal, the eye monitor should be observing a signal centered around 0mV. Are the inputs AC coupled? The down shift could possibly come from a larger magnitude negative voltage swing than positive voltage swing. Before suspecting anything wrong with the eye monitor reading procedure, I'd double-check the circuit first.
Thanks,
Michael
Hi Kawai-san,
The bit transition point should appear at the same point if the same board is used at the same data rate. Changing data rates result in a different time where the bit transition is caught after the delay. For example, compare the eye at 10.3125Gbps v. 1.25Gbps:
10.3125Gbps Bit Transition for EOM, Non-Centered
1.25Gbps Bit Transition for EOM, Non-Centered
If you want to use the GUI (ALP) to show the eye pattern of the device, you are correct in assuming that you will need to interface the PC through USB.
Case 1: I have tried using the DS110DF410EVM to control the registers of another DS110DF410EVM, and had some trouble doing so, because if you plug the mini-USB into the EVM, ALP will not allow you to select the SMBus address.
Case 2: I have found that you can select the target address by connecting the SMBus lines to a SPA dongle instead of a DS110DF410EVM board. You can use the ALP without the EVM as long as you have access to the SCL/SDA/GND lines on the test device and connect them to an interface like the SPA dongle. If they use another microcontroller, then the driver will need to be rebuilt.
Thanks,
Michael
