Uncovering the Challenges of Delivering GigE Over Cat5

Uncovering the Challenges of Delivering GigE Over Cat5
With the enterprise sector looking to corporate LAN environments, interest in Gigabit Ethernet technology is at an all time high. Network interface cards (NICs) are dropping in price, switches are more readily available, and software is stabilizing, making Gigabit Ethernet even more attractive to enterprise customers.

The biggest benefit customers realize in the move to Gigabit Ethernet is a 10X performance increase over existing copper solutions. On the surface, this 10X increase looks like a big win. But, when designers scratch the surface, they quickly realize that achieving this data performance over Category 5 (Cat5) cables designed for Fast Ethernet provides signal-to-noise ratio (SNR), attenuation, and other pesky design headaches.

In this article, we'll uncover some of the headaches developers will face when building Gigabit Ethernet equipment that operate over existing Cat5. Let's start the discussion by looking at SNR margin.

Immunity to noise
SNR margin is a measure of a communications system's immunity to noise. SNR margin is expressed in dB and represents the level of additional noise that the system can tolerate before violating the required bit error rate (BER) of the system.

For example, an SNR margin of 3 dB means that if the noise level were increased by 3 dB, the system would be subject to excessive errors. In other words, SNR margin is a measure of how much additional noise a system can tolerate or how far the system is from not working properly.

SNR margin is a particular problem for Gigabit Ethernet systems. In Gigabit Ethernet designs, the logic levels are increased from three to five while keeping overall voltage constant at 2 V. This results in a reduction in the vertical eye opening by 50%, thus increasing the SNR margin and allowing the system to tolerate less noise (Figure 1). Hence, the noise voltage required to cause a symbol error on a three-level signal is half (or 6 dB lower) than the voltage required to cause a symbol error on a five-level stream. As a result, a three-level signal offers a 6-dB lower SNR margin that emerging five-level designs.

Figure 1: Gigabit Ethernet designs reduce the vertical eye of opening. In this figure, the left chart maps an existing Fast Ethernet eye diagram while the right side denotes a Gigabit Ethernet system.

Correcting for Errors
As stated above, the Gigabit Ethernet standard calls for the use of a five-level coding scheme, dubbed PAM-5. In this scheme, the first four levels are used for data transmission and represent 2-bits of information. The fifth level provides redundancy for use in forward error correction.

In a Gigabit Ethernet design, FEC provides the system with a second level of coding for recovering transmitted signals in the presence of noise. The error correction method used in Gigabit Ethernet combines trellis coding (4D/five state) with Viterbi decoding. It is estimated that this FEC error correction logic enhances the GigE system's SNR margin by up to 6 dB, thus providing GigE systems with the same noise immunity as the three-level Fast Ethernet system.

The FEC capabilities provided by Ethernet are designed to combat three major sources of noise on a Cat5 cable: crosstalk, transmit echo noise and, ambient noise. Let's look at all three a little further.

Crosstalk is defined as unwanted or undesirable signals caused by the electric or magnetic fields of one signal affecting another adjacent signal. Since there are a bundle of four adjacent signals in Gigabit Ethernet transmission, crosstalk becomes a complicated problem to analyze (Figure 2).

Figure 2: Sources of noise in a Gigabit Ethernet system.

Crosstalk is characterized in reference to the transmitter. Near-end crosstalk (NEXT) appears at the output of a wire pair at the transmitter end of the cable. Far-end crosstalk (FEXT), on the other hand, appears at the output of a wire pair at the far end of the cable from the transmitter.

Transmit Echo Noise
Now let's turn to transmit echo noise. Gigabit Ethernet achieves 1-Gbps operation, in part, by transmitting and receiving simultaneously over each of the four twisted-pair cables (250 Mbps equivalent data rate per line). As a direct result of this full-duplex transmission, transmit echo noise is introduced to the system.

In simplified terms, transmit echo noise occurs when the outbound transmit signal reflects (or echos) off the channel (or cable) and creates an unwanted signal at the receiver of the same channel. This echo noise signal interferes with the actual receive signal.

The IEEE Gigabit Ethernet standard defines echo cancellers which attempt to separate the transmit echo from the received signal. This concept of echo also applies to a telephone connection over a single twisted pair. Your phone has built-in echo canceling (much more simplified than Gigabit Ethernet) so you don 't hear your voice reflected back.

To implement bi-directional full-duplex transmission over each of the four Cat5 pairs, the transmitter and receiver are connected to each pair through a directional coupler circuit, known as a hybrid, which separates the outbound transmit signal from the inbound receive signal.

Hybrid networks with good trans-hybrid loss minimize the amount of transmitter signal that is coupled into the receiver but cannot remove all of the transmit echo signal. The residual transmit signal due to the trans-hybrid loss combined with the cable return loss produce the unwanted echo noise.

The magnitude of the reflection, or echo, is proportional to the return loss of the channel. Return loss, usually expressed in dB, may be defined as a measure of the dissimilarity between impedances in metallic transmission lines and loads. Thus, return loss is a measure of the amount of power reflected due to cabling impedance mismatches—the higher the return loss, the greater the cable impedance mismatch.

Ambient Noise
Ambient noise is the final major noise source in a Gigabit Ethernet design. Ambient noise may be defined as unwanted electrical interference on a signal generated by sources in the environment, both near and far. Ambient noise comes in several sources including background white noise and impulse noise generated by power lines and telephone voltages. Interfering wireless signals and alien crosstalk can also create ambient noise.

Due to its random nature, ambient noise cannot be cancelled in the receiver, thus reducing SNR margins. The best defense is a robust system with extra SNR margin. The higher the SNR margin, the less prone the system will be to errors caused by ambient noise.

Attenuation Aches
Clearly, when dealing with any cable attenuation plays a big role in the effective delivery of communications signals. This is especially true in developing Gigabit Ethernet systems that operate over existing Cat5.

Attenuation is defined as the reduction of signal strength during transmission across the cable. Attenuation, measured in dB, increases with increased frequency and distance. If the signal attenuates too much, it becomes unintelligible, and will result in network errors.

The Gigabit Ethernet specification defines a maximum cable distance of 100 meters. Thus, given known attenuation models for Category 5 cable, it may be determined what the minimum transmit amplitude should be to guarantee a certain minimum distinguishable voltage level at the downstream receiver—up to 100 meters away. Based on this logic, the IEEE Gigabit Ethernet standard defines these minimum voltage levels. It is important to note, however, that as the receiver sensitivity is increased, extra cable distance margin is achieved.

Real-world Considerations
Up to this point, we've discussed some of the common problems designers will face when building Gigabit Ethernet systems that operate over Cat5. Now, let's take a look at some real-word considerations that designers may not have yet considered.

To have this discussion, we need to make a few assumptions:

• Standards-compliant Cat5 cabling is used as specified in ISO/IEC 11801:1995 and ANSI/EIA/TIA-568-A (1995) and tested for additional performance parameters specified in Section 40.7 of the IEEE 802.3ab standard.
• Four twisted-pairs of the above-described standards compliant Cat5 cable are available.
• The network supports a BER of less than or equal to 10^-10.
• The horizontal cabling subsystem is no greater than 100 meters as specified in ANSI/EIA/TIA-568-A and ISO/IEC 11801.
• The horizontal cabling subsystem consists of standards-compliant Cat5 patch cords, cables and connecting hardware as specified in ANSI/EIA/TIA-568-A and ISO/IEC 11801.
• The impedance of the Cat5 cable is 100 ohms +/-15%.
• The Gigabit Ethernet systems meet or exceed FCC Class A/CISPR operation.

The environment required by the 1000BASE-T standard for successful transmission of Gigabit Ethernet data exists for the majority of the current installed base of network infrastructure. However, due to the extreme requirements and challenges of the 1000BASE-T standard, as well as the fact that cabling subsystems have been evolving over time, there are many circumstances in which a 1000BASE-T installation will fail. Let's explore some of these cases.

Quality/Existence Not a Given
Like networking standards, cabling standards are constantly changing and evolving to meet the changing and growing needs of the industry. Cabling standards are primarily defined by the ANSI/EIA/TIA-568 specification.

The initial ANSI/EIA/TIA standard was published as recent as 1991 and, at that time, was targeted for 10BASE-T electricals or Cat3 cable with a maximum frequency of 16 MHz. Prior to this specification, cabling was not standardized, and as a result, cabling existed that was sub-Cat3.

In 1995, Cat5 cable was standardized (100 MHz frequency), with the initial driving force being a copper version of the fiber-based FDDI networking protocol, and later, Fast Ethernet. With respect to Gigabit Ethernet performance, it 's important to realize that not all Cat5 cable was/is created equal. Category 5 cable existed prior to the standard being ratified and as a result could result in poorer performance. Additionally, the performance of Cat5 cable is, in part, a function of the chemical compounds used in the manufacture of the cable. It is estimated that from 1994 on, there are 105 different electrical designs of Cat5 cable.

Attenuation Length Doesn't Always Meet Spec
As stated above, the ANSI/EIA/TIA-568-A and ISO/IEC 11801 standards specify that the horizontal cabling subsystem is to be no greater than 100 meters. The horizontal cabling subsystem is defined as the connection between a workstation and the local telecommunications closet and consists of cables, patch cords, and connecting hardware (Figure 3).

Figure 3: Diagram of an ANSI/EIA/TIA horizontal cabling subsystem.

With proper, certified installation it is very likely that the 100-meter maximum distance limit will be guaranteed. However, there are numerous cases where the 100-meter maximum distance limit is exceeded due to uncertified or careless installations, as well as intentional violations. For example, consider a case where a small number of workstations slightly exceed the 100-meter distance limit to the wiring closet. As opposed to incurring the expense of installing an additional intermediary wiring closet, the distance limit may be stretched.

The make up of the cable also plays a role in the performance of a Gigabit Ethernet stream. Some cabling installations will exhibit higher attenuation due to poorer grades of cable or due to imperfections in cabling installations resulting in kinked or distorted cables. The additional attenuation due to these imperfections will decrease the budget for cable length, or conversely, increase the effective cable length (for example, a 100-meter cable will have the attenuation characteristics of a 120 -meter cable).

Cable Installation Problems
Cable installation is also a possible source of network errors, even when the proper standards-defined cable is used. The EIA/TIA-568-A specification also defines the cable installation methods and procedures used by Gigabit Ethernet designers. Cat5 cable requires careful and precise installation to maximize efficiency and system performance. Anything that kinks the cable or disturbs the precise alignment of the pairs inside the cable has the potential to create future performance problems. Some sources of failure are as follows:

• Excessive force used when "pulling " the cables during installation. The EIA/TIA-568-A specification limits the pulling tension to 25 pounds maximum.
• Crushed or pinched cable bundles from sharp edges during installation or from applied weight.
• Crushed cable bundles from support hardware or staple guns.
• Improper termination to connecting hardware. To meet Cat5 cabling standards, the untwisting that takes place where cabling is terminated to connecting hardware should not exceed 13 millimeters, or 0.5 in.
• Loose, worn, or abused wiring can cause the twisted pair to untwist more than necessary. The greater the distance that wires run parallel to one another, the greater likelihood that magnetic fields might cause impedance mismatch or induce current charges.
• Splicing of wires between the telecom closet and outlet locations.

Real-World Considerations Count
As highlighted in the above sections, there are several examples of real-world networking environments that exceed the requirements and assumptions of the 1000BASE-T standard. For the successful market adoption and mass deployment of 1000BASE-T systems, it is critical that Gigabit Ethernet systems not only be compliant to the 1000BASE-T standard, but that they exceed the requirements of the standard enabling the deployment of Gigabit systems in the vast majority of networking environments. In the end, when an end-user or IT personnel purchase and install 1000BASE-T Gigabit Ethernet equipment, it must work properly and transparently. The issues raised in this paper may be addressed and overcome with robust Physical Layer (PHY) technology.

About the Author
Jason Knickerbocker is a product manager at Marvell Semiconductor. Jason holds a BSEE from the University of California, San Diego and can be reached at jasonk@marvell.com.