Christopher Hansen and Professor Stephen Schultz, Electrical and Computer Engineering
Over the last few decades, technological innovations have enabled computers to do things which might once have been considered miracles. Communication systems reach far and wide. With an ever-increasing need for faster communication systems, it becomes necessary to push past the limits of current technology and to innovate new solutions. This project focuses on a commonly used wireline communication technique – “differential signaling”. This method is used within many systems with which most of us are familiar. Ethernet, USB, and PCI-Express are examples of systems which utilize differential signaling. This project seeks to improve upon differential signaling and therefore on communication speeds across wires.
Differential signaling is fairly easy to understand. The basic premise is this: instead of using one wire to transmit a signal, two wires are used. This offers many advantages and some disadvantages. The primary advantage is that it increases the frequency and data rate at which the wires are able to operate. This is because differential signaling offers “noise rejection”. As frequencies go up, noise becomes more of a problem. However, differential signaling is able to counteract that problem enough to be able to achieve frequencies of greater than a gigahertz across two wires. Yet, differential signaling uses two wires to transmit one signal, and therefore is inefficient from a perspective of space required. The amount of power required per piece of data is also greater than that for a one-wire signal.
The method I proposed to study involves the use of three or more wires rather than two wires – “Trifferential Signaling”. By using the same kind of receivers differential signals use, we enable noise to be cancelled, just as with differential signaling. However, instead of only sending one signal at a time, multiple signals are sent simultaneously, and the power-per-unit of data is lower. We aim to send more data with trifferential signaling at the same frequency, thereby keeping the advantages of differential signaling while minimizing the disadvantages.
The funds from the ORCA grant funded the manufacturing of the printed circuit board (PCB) and prototyping supplies. A Digilent Atlys FPGA board was used to generate and compare differential and trifferential (3-wire differential) signals. We designed a custom PCB to connect to the FPGA board as shown in Figure 1.
The Atlys generates differential signals which are sent through the connector which connects the two boards. The differential signals go out through the connector, loop around the PCB, and go back into the Atlys. For the trifferential system, the signals are generated with resistive dividers. The three PCB wires loop back similarly into the Atlys, but fan out in order to enable the subtraction of each wire from each other wire.
This project is ongoing. The data which thus far has been collected, utilizing the above method, shows great promise for trifferential signaling. We measured the performance, in terms of maximum operable frequency, for a two differential pair loopback system versus a 3-wire trifferential loopback system. The differential system was able to achieve a maximum frequency of 111 MHz, for a data rate of 222 megabits per second across four wires. The trifferential system achieved a maximum frequency of 61 MHz, sending two bits at a time, for a data rate of 122 megabits per second across three lines. This achieved the goal of increasing the bandwidth for similar frequencies over differential pairs. The reduced maximum frequency is likely due to the inability of changing the sampling location for the received data with the system used, combined with lack of support with existing PCB layout tools for controlled-impedance three-wire circuits.
Since the manufacturing of the first PCB, I have learned many things to change in order to improve the quality of the trifferential signal. In the PCB in Figure 1, the voltages on some of the trifferential wires unexpectedly attenuated compared to the differential wires. This is due to having too large a resistance in the resistor divider. Lowering this resistance should increase the frequency at which the trifferential system can operate. Additionally, better results would have been achieved had the differential signals been passed through a similar resistor divider. This would have yielded more comparable results because the signals would have had similar properties. I will continue to improve my implementation in order to achieve even better results. Furthermore, due to the timing constraints of the project, only one implementation method was attempted. The primary obstacle in the investigation is choosing the best circuit implementation out of many. More than five distinct implementations were considered, and the one that was chosen was the easiest, quickest, and most cost effective. Other, more robust implementations are already in the works. The results of the first attempt at implementation are pleasing, but leave room for improvement as described above.
With more engineers on the project, a bit more funding, and some time, I believe that we can achieve even more promising results and demonstrate the feasibility of switching to such a system. If the frequency and noise limitations can be overcome through proper design, I believe there will be a strong motivation to switching to the new system, both within cables and on printed circuit boards. By squeezing in more data per wire within the trifferential system, we demonstrate that the bit density can be increased on existing systems. Because the new system uses the same voltage ranges as previous systems but increases the number of bits per unit of data, the new system is greener. This makes the system attractive for large power-consuming data centers and networks.
Being only one person, it is impossible for me to become specialized enough in every area of expertise required to implement the most effective solution. This project requires the expertise of multiple individuals. The ORCA grant enabled me to begin an investigation into trifferential signaling, and I have already learned much.
The initial results are promising and merit further investigation. Given more time, and after building other implementations, I believe we can achieve results that will have a significant impact in industry, in products that we utilize every day, such as USB, Ethernet, HDMI, Lightning, DisplayPort, and PCI-Express. I plan to continue my investigation with the help of a donation by Xilinx which will enable further research with better prototyping tools. I hope to find more individuals who are interested in researching and funding this technology with me in order to enable better cabling and PCB technologies.