Channel Operating Margin (COM) for Serial Link
Channel Operating Margin (COM) is a figure of merit for a passive channel expressed in decibels and is calculated using the ratio of signal amplitude factors to noise amplitude factors. Channel bit rate, insertion loss, return loss, cross-coupling, transmitter and receiver equalization and IC package models are some of the factors applied to determine COM. While it is required for compliance in some applications, COM can also be a valuable part of channel design methodology in general. This example assumes that you have familiarized yourself with the topic page, Channel Operating Margin (COM).
The IEEE 802.3bj 100GBASE specification defines the 100GBASE interface to consist of four channels each operating at 25.78125Gbps. These channel designs can involve PCB only, backplane or copper cables. Signaling is accomplished with either NRZ (Non Return to Zero) or PAM4 (Pulse Amplitude Modulation). Encoding the packets with forward error correction (FEC) is optional but can greatly improve a channel BER (Bit Error Ratio). Testing the compliance of the passive electrical channel to the specification requires it to meet or exceed what is known as COM (or Channel Operating Margin) as measured in decibel units. This document provides information on COM and how to use it within a Serial Link Designer project, having is an interface that operates at 25Giga-Baud per Lane.
COM is a figure of merit derived from the scattering parameters of the passive channel. The overall objective is to give the user insight on the quality of the passive channel design. The calculated metric is related to the ratio of the calculated signal amplitude to its calculated noise amplitude. Channel bit rate, insertion loss, return loss, cross-coupling, transmitter and receiver equalization and IC package models are some of the factors applied to determine this figure of merit. Figure 1 shows a channel model with associated test points. The passive channel referenced is between TP0 and TP5 as documented in IEEE 802.3bj. It is important for the user to keep in mind that COM is required for compliance in some IEEE 802.3 and OIF CEI standarads, but can also be a valuable part of a channel design methodology.
COM Example Project
You can reference the implementation kit COM_25G_BP, which is a Serial Link Designer project consisting of a 25Gbps per-lane design, and can be used as an example to show the procedure for running COM within Serial Link Designer. This can be also applied to various IEEE 802.3 and OIF CEI as well as general channel design methodology.
Signal Integrity Toolbox recommends you to set the Java Heap Memory to at least 8192MB. Use Home Tab -> Preferences -> MATLAB -> General -> Java Heap Memory Current Java Heap: 368MB with 144MB in use. Maximum heap: 1584MB. Heap setting: 1728MB.
The procedures, some examples and tips are given to demonstrate how one may use COM in the analysis of simulation results. Figure 1 is the schematic of the channel design that has been created for analysis.
Figure 1. 25Gbps-per-lane Backplane Schematic Example
The channel design is of a 25Gbps-per-lane backplane with two line cards and high speed connectors. It is a custom 25Gbps interface designed to meet a target 1E-15 bit-error-ratio. Two identical schematic sheets are included in the project with different solution spaces. One is configured to evaluate the effects of loss on COM and the other is to view crosstalk effects. The “Crosstalk” sheet varies aggressor spacing to demonstrate the effects of crosstalk. The “Length” sheet varies the backplane trace length which affects channel loss and thus will affect the channel COM. TX and RX s-parameter package models are included on the schematic sheets. Thus package characteristics will be included in the channel s-parameter models that will be passed into COM. The spreadsheet has been edited to exclude the any package model from the COM calculation as it will be part of the channel model.
Running COM in MATLAB
Step 1: Identify Victim Channel
When the channel design is ready for simulation and the spreadsheet and reported results have been identified, the user can start the process of running COM from within the Serial Link Designer App. The user will need to first identify the victim channel on each of the schematic sheets being simulated. Serial Link Designer requires this such that it will create the appropriate s-parameter files for the victim and any aggressors. To identify the victim channel “Designator Element Properties” must be edited on the schematic sheet. The designator element properties window can be accessed by double clicking on any one of the TX or RX designators on the sheet. Figure 2 shows the example project schematic with victim net identified and the element properties window. The report checkbox in the element properties window should be “checked” for the RX designator of the victim channel only and should be “unchecked” for any other RX designators. The FEXT and NEXT aggressor channels will be automatically determined based on their position on the schematic sheet with respect to the victim. In the case of Figure 2 the RX_Test designator is defined as the victim (See Designator Elements Properties window).
Figure 2: Selecting Victim Channel (Designator Element Properties Window)
Step2: Set Simulation Parameters
It is important that the user set the appropriate simulation parameters in Serial Link Designer prior to running simulations. The extracted s-parameter models for use in COM must have the necessary bandwidth and frequency spacing to get an accurate representation. These parameters are based on the characteristic delay and the channel bit rate. The parameters affected are “Max Output Frequency” and “S Param Frequency Step”.
The “Max Output Frequency” parameter should be at least two times the fundamental, or Nyquist, frequency based on the channel bit rate (NRZ signaling). For example, if the channel simulations are based on 25.78125Gbps, the Nyquist or fundamental frequency would be approximately 12.89GHz. The output frequency should then be set to at least 25.78125GHz.
The “S Param Frequency Step” setting is based on the through path delay of the channel can be used to calculate the step size. The calculation is based on a settling time of three round trip delays for reflected signal energy.
The simulation parameters are set in the Serial Link Designer “Simulation Parameter” window (See Figure 3). To open the Simulation Parameters window, select the pull-down menu SetupàSimulation Parameters.
Figure 3: Simulation Parameters
Step 3: Simulate the Schematic Sheet
The schematic sheet, or sheets, must be simulated so that Serial Link Designer can generate the necessary s-parameter models for COM. Only network analysis has to be run to create the models, but statistical and time domain can be run if desired (See Figure 4).
Figure 4: Simulation Dialog Window
Step 4: Launch COM Interface
After simulations complete COM can be run directly from the GUI. The COM interface can be accessed under “Tools” à“Run COM Interface” (See Figure 5). The Serial Link Designer interface directly invokes MATLAB and the COM application.
Figure 5: Launch COM Analysis
Step 5: Setup COM
Once the MATLAB application starts, the user will be asked to select the COM configuration spreadsheet and the COM code (Figure 6). The spreadsheet being referenced is the one containing the COM parameters that was configured in Step 1. For this example the file is located in the folder “si_lib\COM\” inside your project folder and is in Excel (.xls) format. The spreadsheet can be kept anywhere on a computer or network as long as it is accessible when browsing the system or network. The second file required is the most recent version of the file “com_ieee8023_93a.m” in order to run COM analysis with MATLAB.
Figure 6: Spreadsheet and Code Selection Window for COM simulation.
Step 6: Run COM Script
Click the “Run COM” button. As the code runs the status is reported in the MATLAB Command Window.
Step 7: View COM Results
Once the COM simulation is complete, the Signal Integrity Viewer will open automatically, and the results of COM and the previous simulation results will be loaded. The network, Statistical and Time Domain tabs in the SiViewer all contain the selected results from the COM simulation (See Figure 7).
Figure 7: COM Results Loaded in SiViewer Window
Plotting COM Results
Signal Integrity Viewer offers many capabilities for viewing simulation results. When analyzing a channel design the user may want to look at eye diagrams, loss plots or noise characteristics. Waveform mode allows the user to view data as a function of frequency or time. Some typical results viewed in waveform mode would be the eye diagram of the signal along with its respective bathtub curves and clock PDF (See Figure 8). Another may be an insertion loss versus frequency against a compliance mask as shown in Figure 9.
Figure 8: Statistical Eye Diagram with Bathtub Curves and Clock PDF
Figure 9: Insertion Loss against a Compliance Mask (Black Line)
The channel COM and other results reported by the COM code are given as single data points. The COM result in particular, as being only a single value, makes it easy to determine the pass/fail behavior. However, if one wants to do investigations into the dependent and independent variables of the simulation results, as they pertain to COM, special plotting capability is needed.
The SiViewer has powerful plotting features when “Plots” mode is selected. This mode allows one to uniquely analyze results and create custom plots of virtually any parameter, variable or result from the simulation. It is an invaluable feature for analyzing large numbers of simulations with many variables. Using “Plots” mode gives the user the power to define multiple variables and plot them against each other on the X or Y axis. Figure 10 shows how to access plots tab in the SiViewer tool. Using this advanced visualization technique one can gain greater insight on the channel or system design especially with very large databases. Identifying trends and finding outliers in the results along with custom plots creation can be the key to a successful design.
Figure 10: Signal Integrity Viewer Plots Tab
Investigating COM Results
The example project is set up to demonstrate sweeping backplane trace length and line card aggressor spacing. The first case examined is a sweep of the backplane trace length (“W1” in the solution space) to see how the insertion loss of the channel affects the reported COM value. This is done in the project schematic sheet entitled “Length”. The second case is the variation of the aggressor spacing of two of the PCB traces in the channel (W3 and W7) to observe the effect of coupled noise on the victim channel with respect to the reported COM value. This schematic sheet is named “Crosstalk”. Once Serial Link Designer and COM simulations had finished, the reported COM and statistical BER were compared. The results show an interesting relationship.
As a side note, part of the COM calculation is the determination of optimal equalization settings for the TX and RX with respect to the channel. COM outputs the tap values and from this representative settings were used TX in the statistical simulations. The RX AMI model used in the simulation has an auto adapt feature for both DFE and CTLE. This feature was used in lieu of extracting fixed tap settings from the COM RX adaptation.
Figure 11 is a plot of COM versus the backplane length. In the plot it can be seen that as backplane length (W1) is increased from 8 inches to 20 inches (X-Axis), the COM value decreases from approximately 3.95dB to approximately 2.8dB. The COM 802.3bj compliance requirement for a 100GBASE-KR4 application is 3dB which is marked by the horizontal line on the plot. The data shows that a backplane length up to approximately 18 inches would meet the compliance requirement for COM assuming all other variables in the channel remained constant.
Figure 11: COM versus Channel Insertion Loss
The statistical BER reported by Serial Link Designer for the sweep of backplane length is shown in Figure 12. The BER limit of 1E-12 is marked horizontally on the plot. This plot reveals that backplane lengths up to 15 inches would meet the BER requirement. The plot also reveals some possible resonances in the channel that affect the BER between 8 inches and 12 inches where the BER actually goes down as the length is increased. This behavior is does not exhibit itself in the COM values reported in Figure 11.
Figure 12: BER versus Backplane Length
Figure 13 is a comparison of BER and COM for each case. An interesting observation is that the BER limit is reached at COM values between 3.4dB and 3.3dB. One possibility for this would be that the TX and RX equalization of the IBIS AMI models in the simulation could be better optimized. Another possibility is accounting for the receiver sensitivity and calculating a BER based on different eye contour requirements.
Figure 13: BER versus COM
Figure 14 is a plot of COM versus the aggressor spacing of the trace on either side of the AC coupling capacitors (W3 and W7 on the schematic sheet). The spacing between the victim and each of the aggressor channels was varied from 20mil to 40mil in 5mil increments and is plotted on the X-axis. The horizontal marker represents the COM limit of 3db. The data shows that aggressor spacing of 30mil and greater in the solution space meet the COM limit. Following the trend one could extrapolate that 26-27mil would probably be right at the 3dB line.
Figure 14: COM versus Aggressor Spacing (Line represents COM limit, above the line passes)
Looking at statistical BER versus the aggressor spacing (Figure 15) one can see a very similar trend to Figure 14 where aggressor spacing above 30mil meet the BER requirement of 1E-12. Although the overall results agree the 30mil data point is a slight outlier from the trend. A straight line extrapolation would reveal that 30mil spacing would not meet the target BER.
Figure 15: Channel BER versus Aggressor Spacing (Line represents BER limit, below the line passes)
Lastly, Figure 16 shows BER versus COM for the aggressor spacing variation of W3 and W7. The data is plotted from left (40mil spacing) to right (20mil spacing). The 1E-12 BER limit is marked horizontally and any result below the horizontal marker would meet the requirement. The plot shows that COM and statistical BER agree, however COM results below 3.3dB would not appear to meet the target BER. This is consistent with the findings of the backplane length variation.
Figure 16: Channel BER vs. COM (Horizontal line: BER 1E-12, Vertical line: COM 3dB)
COM returns many other metrics which can be very useful when evaluating performance or diagnosing a problem with the channel design. Going into further detail of these is beyond the scope of this document. So the reader is encouraged to read through the 802.3bj specification and the documentation that is provided with the COM code. These documents show how COM results are calculated and reported which users can utilize those which may be helpful in the design or debug of their channel.
This example provides the information necessary to use the IEEE 802.3 COM application from within the Serial Link Designer App. This document suggests that although a channel meets COM it must still be simulated to determine if the TX and RX can provide enough equalization to meet the target BER. Meeting COM is essential for various IEEE 802.3 and OIF CEI standards compliance and can also provide useful metrics when incorporated in a channel design methodology.
IEEE Standard for Ethernet: Amendment 2: Physical Layer Specifications and Management Parameters for 100 Gb/s Operation Over Backplanes and Copper Cables.
802.3bj Specification: 802.3bj_2014.pdf
COM Quick Guide for April 2014: R. Mellitz (Intel Corp.), Adee Ran (Intel Corp.)
COM Configuration Documentation: config_com_ieee8023_93a_doc.pdf