January 24, 2017
Channel Operating Margin (COM) measurements were first used to address 25 Gb/s channel rates of 100 GbE. They continue to be an essential analysis tool in the emerging 53 Gb/s channel rates of 400 GbE standards for both NRZ and PAM4 signaling, as a means for signal integrity engineers to verify designs. By prescribing minimum values for COM, designers can choose how they optimize signal impairments and equalization schemes while meeting BER (bit error rate) specifications.
While COM offers high-speed designers benefits, it is a complicated, difficult measurement that incorporates many factors. For that reason, the accuracy of the test instruments – especially the vector network analyzer (VNA) – becomes paramount. Understanding COM and the VNA features required to conduct highly accurate measurements on equipment used for high-speed systems is essential to success.
Figure 1 shows the concept of COM. The eye diagram used to measure COM is calculated from measurements of S-parameters, jitter, and noise along with models of serdes (serializer-deserializer) response. The calculated eye diagram includes the insertion loss and return loss of the channel, crosstalk, and random jitter and noise with equalization schemes at both the transmitter and receiver. COM is given by the ratio of the signal amplitude, Asignal, to the vertical eye closure, ANoiseXtalk, defined with respect to the system error rate:
The transfer functions of the channel, its crosstalk aggressors, transmitter output, and receiver input are calculated under the assumption that they are LTI (linear time invariant). Engineers use system transfer function to calculate the response of a single pulse, also known as the single bit response, or SBR(t). In NRZ terms, think of a pulse as a single logic 1 in a long string of 0s; in PAM4, it’s a single S3 in a long string of S0s.
The equalized SBR and all the signal impairments are used to calculate the vertical slice of the eye diagram centered at the time-delay sampling point where the detector error rate (DER) is a minimum. The peak signal amplitude, Asignal, is the signal level and the noise-crosstalk amplitude, AnoiseXtalk, is the vertical eye closure defined with respect to the DER prescribed by the respective standard.
With so many steps in the process, it’s difficult to estimate how the uncertainties of the measured inputs propagate to the uncertainty of COM. When confronted with COM measurements of a given system that are performed by multiple people using different equipment, it is difficult to determine whether the two values are consistent or which is more accurate.
If the measurements have significant disagreement, compare the measured inputs. Pay close attention to the S-parameters because they have a large impact on every step of the COM derivation, except the spectral noise. By measuring S-parameters with a high-performance VNA, such as an Anritsu VectorStar® or Shockline™ model, S-parameters won’t introduce large uncertainties.
The only way to be certain that two measurements disagree is to calculate their total uncertainties. This is achieved by repeating the calculation many times, each with the measured quantities varied within the limits of their individual uncertainties. Every calculation yields a different COM value and the total uncertainty can be calculated from the resulting distribution.
If the measured inputs are consistent but the COM results are not, there is a difference in implementation. Look for a discrepancy by comparing intermediate results of the calculation such as values of the residual ISI, jitter-to-amplitude noise conversion, and crosstalk. If the two calculations seem identical and the measured inputs are consistent, the issue is most likely from the optimization of the equalization parameters.
Once the tools are in place, calculating COM should be fast and efficient compared to a simulation. The purpose of COM is to characterize a channel in a system with a minimally-specified serdes, but COM can also help determine the functionality of a serdes with different channels. COM can help determine the best equalization schemes: the number of FFE or DFE taps and whether transmitter FFE and receiver CTLE are needed. Engineers can estimate the interoperability of different parts, but must remember COM is not a replacement for IBIS-AMI (Input/output Buffer Information Specification-Algorithmic Modeling Interface) models. It cannot verify a design to the degree of a simulation.
The derivation of COM provides diagnostic information through intermediate quantities used in its calculation. These include residual ISI and the interplay of specific transmitter FFE, receiver CTLE and DFE equalization techniques, the impact of spectral noise inside the receiver, and the worst-case impact of crosstalk.
Accurate, Reproducible COM measurements
At data rates of 25+ Gbaud, the accuracy of multiport S-parameter measurements takes on greater importance. Since COM involves so many intermediate calculations, even small S-parameter inaccuracies can lead to large COM discrepancies and days wasted trying to discover the problem.
S-parameters estimated from TDT/TDR measurements lack the bandwidth, dynamic range and noise floor to calculate reliable values of COM. COM measurements require noise floors of –90 dBm and dynamic ranges over 90 dB. It also necessitates ample bandwidth – for NRZ at 28 Gb/s signaling, 42 GHz to cover the 3rd harmonic and 70 GHz for the 5th harmonic and, at 56 Gb/s, 84 GHz for the 3rd harmonic and 140 GHz for the 5th. For PAM4 signaling, 35 GHz bandwidth at 14 Gbaud and 70 GHz bandwidth at 28 Gbaud to reach the 5th harmonics is necessary.
Comparing two inconsistent COM values is a difficult, time-consuming exercise whose result merely indicates whether the calculation was done correctly the first time. The easiest way to get it right is to base calculations on trusted measurements.
Selecting a VNA for COM measurements
In selecting a VNA there are several important parameters. Table 1 shows typical performance abilities for three models – the ultra-high performance VectorStar VNA and more economical ShockLine model. VNAs measure the device performance one calibrated frequency at a time in steps from the low frequency limit up to the high frequency limit. The low frequency limit is important in assuring that frequency domain measurements can be accurately transformed to the time domain. Since VNA S-parameters are sets of discrete measurements, it’s also important to keep the frequency step size as small as possible; which corresponds to many measurement points.
The combination of unique low-frequency coverage with up to 100,000 measurement points make VectorStar VNAs all but immune to aliasing and causality problems. The more economical ShockLine VNAs also have excellent low frequency coverage and up to 20,000 measurement points. Precise S-parameter measurements also requires accurate VNA calibration and the ability to de-embed the test fixture.
To learn much more on COM measurements and the test tools necessary to make them, download this newly published white paper.