ECN Wed, 12/19/2012 - 11:45am Comments by Clark Kinnaird, Industrial InterfaceSystems Engineer, Texas Instruments
Differential signaling is used for noise immunity inEthernet, RS-485, CAN and USB. In ideal cases, all common-mode noise isrejected. In real-world applications, there are several design techniques andcomponent parameters to consider in order to keep the data flowing with highconfidence. Differential signaling is used in most interfaces, which sendsdigital information over cables. Although requiring two signal wires rather thanone, differential signals are much more immune to noise than single-endedsignaling.
The basics of differential signaling are well-known, taking advantage of thenoise rejection which affects both signal wires equally. This is illustrated inFigure 1, where a balanced differential signal is transmitted on two twistedsignal wires (twisted-pair).
Figure 1. Typicaldifferential signal chain with simplified model of electrical noise coupling
Electrical noise from the environment affects both wires equally, such that the received signals on A and B are:
VA = + ½ VSIGNAL + VNOISE
VB = - ½ VSIGNAL + VNOISE
so that the differential voltage signal is:
VA – VB = VSIGNAL
Figure 2. When noise isreceived equally on A (purple) and B (blue) signal wires, it is rejected in thedifferentially-received signal (black = A-B).
Popular electricalstandards such as USB, Ethernet, RS-485 and CAN use differential signaling andbalanced twisted-pair media to provide reliable high-speed communication.
In practice, designers should keep in mind that no real system has the idealperformance of a theoretical model. There are several key sources of errors andnoise that should be considered.
Line-to-line impedance imbalance
Balanced signal wires are critical to the noise immunity of differentialsignaling. Twisted-pair cables specify the level of imbalance allowed. Forexample, at low-frequencies CAT 6A is specified with 40 dB of transverseconversion loss. This means that a 1V transient coupled to both signal lines(common-mode) creates only 10 mV of differential-mode signal. Lower grades ofcable allow higher fractions of common-mode to differential-mode conversion.
Imbalance in the differential path can be caused by components added to provideprotection against transients. For example, transient voltage suppression (TVS)components are sometimes prescribed as a means to prevent damage due toelectrostatic discharge, voltage surges, or electrical bursts. Designers shouldcheck the matching characteristics of these components to ensure that eachdifferential line is affected equally.
Figure 3. When noise isreceived unequally due to unequal coupling or unbalanced impedance, a fractionof the noise appears on the differentially-received signal
Other potential contributors to imbalance between the differential linesinclude board traces and connectors. In both cases, the impedance and matchingmay depend on the frequency being considered. Designers should consider thefrequency content of the intended signaling and expected electrical noiseenvironment.
Transmission line length
Inequality of the signal line lengths is another source of differential-mode tocommon-mode noise conversion. If noise is equally coupled to two perfectlybalanced lines, but the noise signals reach the differential receiver atslightly different times, this is seen as non-zero differential noise. Thisbecomes a significant issue as the frequency bandwidth increases.
VA = + ½ VSIGNAL + VNOISE= + ½ VSIGNAL + Asin [t]
VB = - ½ VSIGNAL + VNOISE’ = - ½ VSIGNAL+ Asin [(t+t)]
VA - VB = VSIGNAL + A {sin [t] -sin [(t+t)] }
When t correspondsto a 180-degree shift in the noise frequency, this causes the worst-casecondition where the resulting differential noise has twice the amplitude as theoriginal common-mode noise. But even small phase-shifts can cause significantfractions of common-mode to differential-mode conversion. For instance, a phaseshift of one-tenth radian (about six degrees) causes about 10 percent of thecommon-mode noise to appear as differential signal.
Figure 4. Whenequally-coupled noise reaches the receiver with a time-shift between the twosignals, due to difference in the lengths of the transmission lines, theresulting noise appears on the differentially-received signal
Taking a concrete example,assume the differential receiver for a USB 2.0 device has a bandwidth of atleast 1 GHz to receive 480 Mbps data. If 1 GHz noise is coupled to thedifferential lines, a difference in length of 3 mm corresponds to about 15 picosecondsof time shift, which is about one-tenth radian of phase-shift between the twolines. This can cause 10 percent of the incident 1 GHz noise to appear asdifferential voltage, which could cause unintended receiver state switching.For lower-frequency standards, such as RS-485 or CAN, the receiver bandwidth,and therefore the sensitivity to line-length inequality, is correspondinglyreduced.
Reduced immunity during idle and cross-over times
When the bus lines are being actively driven to a valid logic state, the driveroutput level has significantly more amplitude than the receiver thresholdlevels. This margin assures that the transmitted state is received correctly,even if the signal is attenuated by electrical losses and corrupted with some levelof differential noise.
However, when no driver is actively transmitting, the bus lines are moresusceptible to corruption by induced differential noise. This samesusceptibility occurs during transition from one valid state to another, as thedifferential signal enters the neighborhood of the receiver threshold. Ineither of these cases, relatively small amplitudes of differential noise canmomentarily cause unintended receiver transitions from one output state toanother. During these critical times, receiver hysteresis provides a measure ofnoise immunity.
Hysteresis improves noise immunity
Receiver threshold hysteresis reduces the sensitivity of differential receiversto electrical noise on the signal lines. The amount of separation between thethreshold levels must be controlled so that the overall sensitivity of thereceiver still meets the standard’s requirements. Therefore, receiverhysteresis (usually measured in mV) is an indicator of the differential noiseimmunity of any particular transceiver or PHY. Designers should consider thereceiver hysteresis when concerned with noisy environments.
Figure 5. Receiverhysteresis improves noise immunity.
In summary, system designers should evaluateseveral sources of potential problems in their differential networks. Theseinclude cable, connectors, and protection devices, as well as the transceiveritself. The impedance tolerance for each component on the differential signallines should be kept within a total imbalance budget. The transmission linelength of each differential signal must be equal within a small fraction of theshortest wavelength of interest. Board layout and connector pin arrangement areimportant, especially for high-frequency networks. Finally, choose differentialreceivers with more hysteresis for high-noise applications. Taking these stepscan help ensure reliable communications even in applications with noisyenvironments.