As Ethernet and OIF Standards groups wrestle with the technical feasibility of 224 Gb/s (Gbps) PAM4 signaling in order to standardize electro-optical requirements of data center communication, a particular challenge of measurement verification is beginning to emerge. Previous generations of standardization have set a bandwidth minimum target as three quarters of the baud rate. This paper, presented at DesignCon 2024, explores what is meant by bandwidth during the standardization process, the implications of test and verification attached to certain bandwidth requirements, as well as differences between acquisition range, band limited filters, and s-parameters for time domain processing.
Sensitivities of 1.0 mm RF connector mechanical tolerances on measurement to simulation comparison are presented. Channel level performance metrics with the trending of current channel compliance methods are compared. The paper reviews the challenges and potential mitigations ahead.
Experiment Specifics
The work details two case studies where adevice under test (DUT) is configured with two RF connector types (1.0 mm and 1.85 mm) in order to make comparisons between the frequency domain and time domain. Additionally, in the case of the second DUT, 224 Gbps traffic is passed to determine if any impact is noticed between the RF connector types.
The details of each case study are described and include case no.1 as a stripline DUT populated with either a 1.0 mm or 1.85 mm vertical, compression mount RF connector, as well as case no. 2 with a ganged, cabled RF connector attached to a PCB that emulates a 20-dB channel.
Frequency domain (FD) data is collected using a 4 port 110 GHz vector network analyzer (VNA). There is also a detailed discussion around a key FD figure of merit for RF connectors and coaxial cables: cutoff frequency (f c) of propagating higher order modes.
The analysis uses time domain (TD) metrics to compare performance between 1.0 mm and 1.85 mm, where choices are vertical eye closure (VEC) and 12 Edge Jitter. TD performance data is captured using a high-speed oscilloscope with receiver processing capability. A 224 Gbps PAM4 capable arbitrary pattern generator produces a PRBS13Q data stream which is sent through the device under test (DUT) to the oscilloscope. The oscilloscope was configured with equalization and filtering which are expected to be similar to what is deployed in 224 Gbps PAM4 devices.
Figure 1. Lab setup for time domain measurements. High-speed digital electrical data transmission has its basis in electromagnetic wave theory. At 224 Gbps PAM4, the baud rate is 112 GBd and the Nyquist frequency is 56 GHz. The pulse width is 9 ps which in a material with permittivity of approximately 3 means that 1 pulse is 1.5 mm. The paper goes on to explore what new phenomena need to be addressed and how this affects VEC and jitter measurement.
Conclusions and Future Work
The conclusions about 1.0 mm applicability seem to be somewhat mixed. Since 224 Gbps PAM4 standards are trending toward using receiver filters with a cutoff filter around Nyquist frequency either 1.0 mm or 1.85 mm connector seem acceptable for receiver compliance or VEC testing. Since the requirements for bandwidth when measuring jitter has been higher in the past, 1.0 mm connectors may be appropriate for those device characterizations. A note of caution has emerged from this work. Although 1.0 mm RF connectors promise higher frequency domain fidelity in the range beyond the rating of 1.85 mm RF connectors, there is likely to be additional usability constraints, increased costs, as well as manufacturing penalties.
There are several items left for future work. Both the VNA and oscilloscope mating required a metrology grade adapter to mate with the 1.85 mm RF connectors. The the impact of a lower bandwidth pattern generator or oscilloscope that would remove the need for these adapters was not investigated.
The channel emulation DUT was populated with RF086 coaxial cable for both the 1.0 mm and 1.85 mm configurations which means that the propagation of higher order modes was still possible. A comparison at the same DUT loss on the smaller diameter RF047 coaxial cable would have been interesting to pursue. But, due to time constraints and reduction in cable reach to achieve similar insertion loss, this is left as future work.
This paper focused on the time domain metrics as would be employed for host or receiver compliance. The impact of RF connector type and receiver filter cutoff for transmitter devices would be an interesting future consideration. Advanced receiver techniques such as maximum likelihood sequence estimation are becoming more common and are potentially required for 224 Gbps PAM4 long reach electrical channels. Exploring the impact on reducing the receiver bandwidth filter cutoff when these techniques are employed could further help inform the sufficiency of channel bandwidth requirements
The paper referenced here received the Best Paper Award at DesignCon 2024. To read the entire DesignCon 2024 paper, download the PDF.