Guard Against EMI For 10GBaseT And Beyond

Fig 1. Traces of EMI noise in the 460-MHz range are captured by a high-speed digital scope at the chip side of the transformer on the medium dependent interface (MDI) of a 10GBaseT receiver. The noise introduced by walkie-talkies operating in the Family Radio Service (FRS) and General Mobile Radio Service (GMRS) public bands can be as much as 40 dB above the noise sensitivity of 10GBaseT receivers.

Fig 2. Applied Micro’s Triveni Ethernet PHY interfaces come in dual and quad versions. The APM96895, which is a quad version in a 27- by 27-mm package, consumes 3.5 W and can produce 10GbE over up to 100 meters of CAT6a UTP.

The 10GBaseT standard brings significant performance and power efficiency benefits to the data-center, enterprise, telco, and even small-to-medium business (SMB) market segments. However, increasing the data rate from 1 to 10 Gbits/s on unshielded twisted pair (UTP) introduces non-trivial high-frequency noise and electromagnetic-interference (EMI) issues.

UTP copper-based cabling is the ubiquitous medium for network infrastructure given its low cost. Concerns have emerged, though, regarding the reliability of UTP and whether it can meet the bit-error-rate (BER) and distance requirements of 10GBaseT. Specifically, the higher signaling frequency required at 10 Gbits/s, coupled with bandwidth that overlaps the UHF and VHF bands, exceeds the capability of traditional passive noise-cancellation techniques to restore sufficient signal integrity at the receiver.

The UTP Dilemma

Network administrators can choose from three cabling options when introducing 10GBaseT to the data center or enterprise: optical fiber and two copper-based cables, shielded twisted pair (STP), and UTP (see the table). Optical fiber provides the best signal integrity and performance. But when total cost of ownership (TCO) including installation and maintenance is considered, fiber costs substantially more than copper-based cabling.

While optical fiber itself is inexpensive, the manufacture of optical transceivers requires a more expensive process technology than CMOS. Also, a trained technician must install fiber to ensure precise alignment of the laser and fiber. In addition, optical cabling is more vulnerable than copper cabling, limiting how and where it can be installed, as well as potentially requiring more frequent maintenance.

Copper cabling has the advantage of being simple to install and robust enough to run in places where fiber would be too fragile. Nonetheless, while STP cabling offers excellent noise immunity, it’s not widely used in the U.S. due to its substantially greater cost compared to UTP and difficulties with grounding.

Of the three types of cabling, UTP is the most susceptible to interference. Despite this, it has had a tenacious hold on the market during Ethernet’s evolution because it’s the least expensive to install and maintain. UTP’s lasting presence in the communications industry can be traced to three primary factors:

• Advanced signal processing: As data rates increase, communication system theory and advanced digital-signal-processing (DSP) technology progress at a pace that enables physical-layer (PHY) manufacturers to maintain sufficient signal margin to compensate for UTP-incurred losses.
• CMOS analog improvements: Thanks to innovations in analog design techniques, advanced signal-processing techniques can be implemented efficiently in CMOS.
• Low cost: CMOS process technology follows Moore’s Law, leading to predictable cost reductions in high-speed PHY silicon. Such silicon thus can deliver the reliability and performance required to support next-generation Ethernet using UTP.

There’s tremendous momentum behind UTP to continue as the medium of choice for 10GBaseT technology. UTP significantly lowers the barrier to entry for 10GBaseT compared to fiber and STP, and the availability of reliable UTP-based equipment is expected to catalyze adoption of 10-Gigabit Ethernet (10GbE). As a result, many key players in the Ethernet market are invested in ensuring that UTP meets the reliability and performance requirements of the enterprise and data center.

Greater Bandwidth Means Greater Interference

RF interference isn’t a new problem for wired communications. In the 1990s, the 802.3 and DSL standards groups conducted studies to determine the impact of RF interference on throughput and reliability. They also evaluated different ways to add more noise margin to designs to compensate for RF interference.

Various signal-processing techniques can help preserve noise margin. First, equalization techniques like Tomlinson-Harashima Precoding (THP) remove any inter-symbol interference (ISI). Second, most non-alien crosstalks between twisted pairs in the same cable can be removed using traditional adaptive filters. Finally, advanced forward-error-correcting (FEC) codes such as low-density parity checking (LDPC) can close the signal-to-noise ratio (SNR) gap within the Shannon limit.

In addition to those techniques, 10-Gbit/s transmission over UTP could not happen without a substantial increase in noise sensitivity on the analog side. Implicit to this statement is lower background noise in the transmission media, such as cable infrastructure, RJ-45 connectors, transformers, and traces on the board.

Most problematic for 10GBaseT system designers is that the standard increases operation up to the 400-MHz band. Thus, many of the UHF and VHF bands used for various wireless communication devices, including emergency and police radios, are either in-band with 10GBaseT communications or close enough to cause interference.

For example, the EMI generated from a walkie-talkie operating in the family and general mobile radio service (FRS/GMRS) band can be as much as 40 dB above the noise margin of the receiver (Fig. 1). Since this noise is in-band, a simple filter isn’t sufficient to remove it.

Notching out the shared frequencies will result in lost bandwidth for the 10GbE connection. This loss would be significant enough to make the link non-compliant when running the cable at up to 100-meter lengths. In addition, network administrators would deem it unacceptable to allow data-center reliability to become vulnerable to the presence of a public band device.

Active Vs. Passive Noise Cancellation

Each stage in Ethernet’s evolution has required new signal-processing techniques to address increasing noise and interference. The move to 10GbE is no different. Achieving reliable operation of 10GbE over UTP requires improvements across all components in the signal path.

For example, moving to UTP-CAT6a copper cabling will reduce insertion losses and boost signal bandwidth. The transceiver analog front end (AFE) must offer improved sensitivity to counteract the lower noise margin. In addition, the PHY must employ various advanced noise-cancellation algorithms to remove inter-symbol interference (ISI) and non-alien crosstalk. Furthermore, interference from overlapping public bands should be removed without giving up valuable bandwidth by leveraging advanced signal-processing techniques.

Traditionally, Ethernet has employed passive noise-cancellation techniques to eliminate noise. They work very well if the interference is completely out of band with respect to the transmission. Eliminating EMI from in-band RF transmitters, however, can’t be achieved through simple filtering, because it would reduce the overall bandwidth. Also, the random/data-based signature of RF interference must be considered. Effectively canceling in-band interference while maintaining sufficient noise margin clearly requires advanced signal-processing techniques.

Active noise cancellation works by reconstructing the EMI source using whatever information can be derived and subtracting it from the received signal. Because EMI is always changing, the system needs to dynamically follow the noise signal’s changes in frequency, amplitude, and phase variation to estimate the signal’s next change. More accurate EMI tracking performance translates into more accurate noise removal. Comparing the estimated signal with the received signal, then, will dynamically refine the estimates.

Given the fast variations in parameters of the interference signals, it’s crucial to detect and cancel EMI in a short period of time. As a consequence, active noise cancellation requires more advanced DSP capabilities than passive noise cancellation or simple filtering techniques.

Numerous active noise-cancellation techniques are available for rapidly detecting and cancelling RF interference: time-domain pulse templates, modulation, frame structure, and the frequency-domain band information. Manufacturers of 10GBaseT PHY offer a variety of active noise-cancellation technologies integrated directly into the PHY that dynamically track and cancel EMI sufficiently. This, in turn, allows UTP cabling to meet Ethernet BER requirements across the required cable lengths. One example of an Ethernet device using DSP active noise-cancellation techniques is Applied Micro’s Triveni APM9689x family (Fig. 2).

Standards Of Confidence

The Ethernet industry has to recognize that existing noise-immunity standards were originally created for transmission at much lower speeds and, therefore, must be adjusted for 10GbE. Most immunity standards are based on simplistic analog transceivers. As such, they only test the performance of simple, passive cancellation techniques using synthetic sources of EMI. These techniques overlook key features that are fundamental to active noise cancelling.

As they stand, these tests will show whether or not 10GBaseT equipment operates reliably in a compliancy lab. Beyond that, though, they offer little indication of equipment performance when exposed to real-world operating conditions and RF interference.

Since 10GbE uses a higher frequency, noise margin drops significantly and active noise cancellation becomes an essential component for reliable operation when using UTP cabling. Scenarios must be created and adopted that define practical sources of EMI with clear specification on time- and frequency-domain templates to confirm the reliability of links in the presence of EMI. In addition, testing needs to verify that equipment offers continuous operation without long bursts of errors. Finally, test scenarios have to outline product requirements for compliance and indicate the types of EMI sources being tested against.

A successful, rapid adoption of 10GbE depends on thorough testing with transparency of test parameters. In addition to establishing standards for reliable operation, EMI susceptibility testing will boost market confidence in the mass deployment of UTP-based 10GBaseT. It will also enable network administrators to easily compare the noise immunity of products from different vendors and ensure that the equipment they purchase meets the reliability needs of their organization.