INTEGRATED COUPLER-MOUNTED NON-CONTACT ETHERNET BRIDGE

CROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims priority to and the benefit of U.S. Provisional Patent Application Serial No. 60/980,528, filed on October 17, 2007, which is hereby incorporated herein by reference in its entirety.

TECHNICAL FIELD

[0002] The invention generally relates to the field of contactless data signal coupling and, more specifically, to the field of contactless data signal coupling mechanisms optimized for wide band communications over an air gap.

BACKGROUND

[0003] Existing systems use four encapsulated transducers, mounted in electrical coupler heads, to pass electrical signals between two mechanically coupled rail vehicles via a non- contact link. A first pair of transducers is mounted on a first coupler head of the first of the two railcars joined by the non-contact link. The first pair of transducers includes a sender and a receiver, each of which includes active electronics. A second pair of transducers, identical to the first pair, is mounted on a second coupler head of the second of the two railcars in such a way that each sender faces a receiver on the opposite car. For each facing pair, the proximity of the two opposed transducers is such that a baseband Ethernet signal is coupled magnetically from sender to receiver in a near-field transmission mode. [0004] The sender and receiver on each coupler are connected by a shielded cable to a trainline, or segment, interface unit on-board the rail vehicle. The interface unit supplies power to the transducers and conditions the transmitted and received signals to be compatible with a standard Ethernet physical layer device (PHY). Conditioning functions include impedance matching, amplification, equalization, and automatic gain control (AGC). Multilevel transition 3 (MLT3) Ethernet data from a receiving transducer is treated as an analog signal up to the point of presentation to an integrated circuit Ethernet switch

(hereinafter a "switch chip") contained within the interface unit. The switch chip provides a final standardized interface to an external local area network (LAN).

SUMMARY

[0005] Briefly described, and in accordance with exemplary embodiments thereof, the present invention relates to an integrated coupler-mounted non-contact Ethernet bridge unit for providing full-duplex baseband data communications over two adjacent parallel paths, and which includes a single housing, a first magnetic coupling means for providing data communication in a first direction, and a second magnetic coupling means for providing data communication in an opposite direction to the first direction. The first magnetic coupling means and the second magnetic coupling means are mounted within the single housing. The integrated Ethernet bridge unit may include such advantages as greater reliability than ohmic contacts, reduced signal corruption caused by the poorly controlled wiring impedances typical of standard coupler connections, and superior immunity to dirt, water, ice, and contaminants, as compared with both ohmic and radio-frequency connections. [0006] One or more of the following features may be included. The coupler-mounted transducers, along with the interface unit that formerly resided within the interior of the rail vehicle, may be integrated into a single integrated, coupler-mounted, non-contact, Ethernet bridge unit (hereinafter "bridge unit"). The integration of the bridge unit into one housing advantageously provides space to make the magnetics a little larger compared to known magnetics designs, thereby allowing a gap between couplers to be larger for a given signal-to- noise ratio and bit error rate.

[0007] An automated self-test capability, including a means for communicating with a monitoring and diagnostic system (MDS) on-board the train, may be included. The data transmission windings may be incorporated into the main printed circuit board of the bridge unit. The existing ferrite polepiece may be replaced with a pair of rectangular ferrite bars molded or cemented into a protective plastic cover that seals each magnetic aperture.

[0008] Adaptive equalization may be added to improve bit error rate. A low- frequency, low data rate, serial auxiliary communications channel, superimposed on the magnetically coupled Ethernet signal, may be added to facilitate adaptive equalization, synchronization handshaking,

and diagnostic information exchange between the bridge units on adjoining railcars. The auxiliary communications channel between two bridge units is created by means of an amplitude modulation envelope applied to the MLT3 Ethernet waveform. An adaptive equalization loop may provide continuous correction for variations in the frequency response of the magnetics.

[0009] Two independent 100 Mbit networks may be provided. A bridge unit, one instance of which is mounted on each of the two couplers joining a pair of rail cars, may communicate data at 100 Mbits/sec between the cars by means of non-contact inductive coupling of an MLT3 baseband signal. [0010] Half of a near-field magnetic data coupling structure may be located in each of two communicating bridge units. Together, the two halves may form an air-gap transformer for the transfer of Ethernet data between the bridge units. This structure may be implemented twice, once for each direction of data flow. A pair of ferrite polepieces, one located above the plane of the circuit board containing the transmit windings and the other an equal distance below the circuit board, may be situated to facilitate coupling with the receiving half-transformer on the second bridge unit.

[0011] A wideband half-transformer, including a two-layer printed-circuit winding, may be located near one edge of the circuit board inside the bridge unit. The wideband half- transformer may transmit or receive data to or from a receiving or transmitting half- transformer on a second bridge unit proximate the first. A second two-layer winding, which may also be part of the transmitting or receiving half-transformer, may provide a means for looping the transmit or received signal belonging to a given bridge unit back to the receive or transmit half-transformer on the same bridge unit. A signal induced in this winding may be coupled to the receive or transmit half-transformer when the bridge unit is in self-test mode, and may be ignored during normal operation. A selective blocking of flux in certain areas may reduce leakage inductance in the transmit or receive windings. This blocking effect may be mediated by eddy currents flowing in localized copper planes on the printed circuit board. Insertion of a terminating resistor in the winding current path at the geometric center of the winding structure may reduce parasitic transmission line resonance in each two-layer transmit or receive winding.

[0012] A switch chip, which is an integral part of the bridge unit, may provide a standard copper 100Base-TX interface to an external Ethernet LAN. An AGC loop may maintain the Ethernet signal presented to receive terminals of the switch chip at a fixed peak-to-peak amplitude. A secondary function of the AGC loop may be stripping a deliberately imposed amplitude modulation (AM) envelope from the Ethernet signal presented to the switch chip. A tertiary function of the AGC loop may be extracting the deliberately imposed amplitude modulation envelope for further processing. This envelope may include a sinusoidal subcarrier impressed with digital data.

[0013] Data may be impressed on the auxiliary communications subcarrier by multiplication of the subcarrier by a quadrature amplitude modulation (QAM) envelope. Data transported by the auxiliary communications subcarrier may be recovered by a QAM demodulator.

[0014] A transmit frequency analysis filter and accompanying root mean square (RMS) detector may measure the spectral energy of the transmitted and/or received Ethernet signal across each of two overlapping frequency bands. The first band may include only low frequencies and the second may include both low and medium frequencies. The use of the spectral energy measurement spanning both low and medium frequencies as a reference level permits ratiometric compensation for component tolerances in the transmit and in the receive frequency analysis filters and for variations in the sensitivity of the respective RMS detectors.

[0015] The exchange of spectral analysis results between two bridge units by means of the auxiliary communications channel. The transmitted result in each case includes the low- frequency energy content of the Ethernet signal as measured at the originating source.

[0016] A spectral evaluation mechanism, which may operate on the receive side of the bridge unit, may compare the measured low-frequency energy content of the magnetically coupled incoming Ethernet signal with the low-frequency content measured at the originating transmitter, the latter information being conveyed to the receiving bridge unit over the auxiliary communications channel. The spectral evaluation mechanism may operate in the bridge unit at each end of the magnetically coupled data link. Each bridge unit may act as both a transmitter and a receiver of Ethernet data.

[0017] A continuously adjustable equalization filter, which is located on the receive side of each bridge unit and which, when properly controlled, may compensate for the frequency distortion of the Ethernet signal introduced by the magnetic coupling hardware.

[0018] A servo loop, which resides on the receive side of the bridge unit, may use the result of the spectral evaluation to adjust the continuously variable equalization filter. The goal of such adjustment may be to minimize the difference between the low- frequency energy content of the Ethernet signal as measured at the receive input terminals of the switch chip, and the low-frequency energy content of the Ethernet signal as measured at the point of origin (i.e., the "distant" transmitter). [0019] A governing intelligence within the bridge unit includes, in an exemplary embodiment, a digital signal processor (DSP), that permits the various control and communications strategies (such as AGe, adaptive equalization and the auxiliary communications channel) to be implemented at least partly in software. Nonvolatile memory may be within, or closely linked to, the governing intelligence. Such memory may be capable of storing configuration data, such as, for example, a vehicle ID, for consist enumeration. A supervisory serial port may permit the governing intelligence within the bridge unit to communicate with external diagnostic and installation setup equipment.

[0020] The bridge unit may tunnel data from the supervisory serial port through the auxiliary communications channel. The auxiliary communications channel may be a non-Ethernet data link between two bridge units, for coordinating certain internal functions. By means of the supervisory serial port, the auxiliary communications channel may be made available for concurrent use by external non-Ethernet equipment. The supervisory serial port may be used for the download of configuration data to nonvolatile memory and for the collection of diagnostic data.

BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Embodiments of the present invention are illustrated by way of example and are not limited by the accompanying figures, in which like references indicate similar elements. Elements in the figures are illustrated for simplicity and clarity and have not necessarily been drawn to scale.

[0022] FIG. 1 is a perspective view of an integrated, coupler-mounted, non-contact, Ethernet bridge unit and a housing therefor, in accordance with the invention;

[0023] FIG. 2 illustrates a side view of the internal construction of the integrated, coupler- mounted, non-contact, Ethernet bridge unit; [0024] FIG. 3 illustrates the integrated, coupler-mounted, non-contact, Ethernet bridge unit mounted atop an electrical head of a railcar coupler;

[0025] FIG. 4 illustrates two integrated, coupler-mounted, non-contact, Ethernet bridge units mounted atop an electrical head of a railcar coupler;

[0026] FIG. 5 is a simplified electrical functional block diagram of the integrated, coupler- mounted, non-contact, Ethernet bridge unit;

[0027] FIG. 6 is a plan view of the circuit arrangement on a printed circuit board of the integrated, coupler-mounted, non-contact, Ethernet bridge unit;

[0028] FIG. 7 illustrates several views of a coupler face bar without and with magnetics;

[0029] FIG. 8 is a top view of a portion of the printer circuit board, illustrating details of the windings of the magnetics;

[0030] FlG. 9 is a cross-sectional view of a portion of the printer circuit board, illustrating details of the windings of the magnetics;

[0031] FIG. 10 illustrates current flow superimposed on the top view of the portion of the printer circuit board shown in FIG. 8; [0032] FIG. 1 1 is a simplified software data flow diagram for a digital signal processor, in accordance with embodiments of the invention;

[0033| FIG. 12 illustrates an idealized Ethernet waveform;

[0034] FIG. 13 illustrates an MLT-3 transition diagram;

[0035] FIG. 14 illustrates a waveform demonstrating a minimum MLT-3 frequency;

[0036] FIG. 15 depicts a wideband pulse transformer;

[0037] FIG. 16 illustrates a graph of a signal featuring both resonance distortion and frequency distortion;

[0038] FIG. 17 illustrates an example of a transmission line transmitting a bundle of charge;

[0039] FIG. 18 illustrates a pipe and air piston;

[0040] FIG. 19 illustrates a shorted transmission line;

[0041] FIG. 20 illustrates a matched transmission line; [0042] FIG. 21 illustrates a mismatched transmission line;

[0043] FIG. 22 illustrates a system featuring a single-layer solenoid;

[0044] FIG. 23 illustrates a two-layer printed spiral winding;

[0045] FIG. 24 illustrates a two-layer printed spiral winding with a terminating resistor;

[0046] FIG. 25 illustrates a uniform transmission line; [0047] FIG. 26 illustrates a transmission line of arbitrary geometry;

[0048] FIG. 27 illustrates an idealized version of a primary winding structure;

[0049] FIGS. 28-31 illustrate transformer equivalent circuits;

[0050] FIGS. 32-34 illustrate the equalized admittance of the transformer;

[0051] FIGS. 35-36 illustrate receiver waveforms; [0052] FIG. 37 illustrates the circuit topology of a two-stage L-section filter;

[0053] FlG. 38 illustrates an equivalent circuit of the L-section filter; and

[0054] FIGS. 39-40 illustrate a step response before and after optimization.

DETAILED DESCRIPTION

[0055] Disclosed herein are systems and methods for non-contact transmission of 100- megabit Ethernet data between two railcars joined by an electromechanical coupler. In accordance with embodiments of the invention, a baseband signal is magnetically communicated across an air gap formed by the space between the two electrical coupler heads. The foregoing and other features and advantages of the invention will be apparent from the following detailed description taken in conjunction with the accompanying drawings. It should be understood that these embodiments are only examples of the many advantageous uses of the teachings herein. In general, statements made in the specification of the present application do not necessarily limit any of the various claimed inventions. Moreover, some statements may apply to some inventive features but not to others. In general, unless otherwise indicated, singular elements may be in the plural and vice versa with no loss of generality.

[0056] As required, detailed embodiments of the present invention are disclosed herein; it is to be understood, however, that the disclosed embodiments are merely examples of the invention, which can be embodied in various forms. Therefore, specific structural and functional details disclosed herein are not to be interpreted as limiting, but merely as a basis for the claims and as a representative basis for teaching one skilled in the art to variously employ the present invention in virtually any appropriately detailed structure. Furthermore, the terms and phrases used herein are not intended to be limiting; but rather, to provide an understandable description of the invention.

[0057] This subject matter is also described in the following patent applications, which are hereby fully incorporated herein by reference: PCT application No. PCT/US2006/026672, entitled CONTACTLESS DATA COMMUNICATIONS COUPLER, filed July 7, 2006; PCT application No. PCT/US2007/060198, entitled CONTACTLESS DATA COMMUNICATION COUPLER, filed January 6, 2007; and Provisional U.S. Patent Application Serial No. 60/954,386, entitled NON-CONTACT ETHERNET LINK WITH BIDIRECTIONAL TRANSDUCER, filed August 7, 2007.

Packaging

[0058] FIG. 1 illustrates an exemplary embodiment of the general appearance of an integrated, coupler-mounted, non-contact, Ethernet bridge unit (hereinafter "bridge unit") 100, including a housing 102 therefor. An Ethernet connector 104 may be included on the rear of the bridge unit 100, and receiver magnetics 106 and sender magnetics 108 may be included on the front of the bridge unit 100. FIG. 2 illustrates a cross-sectional view 200 of the interior of the bridge unit 100. The bridge unit 100 may contain a single printed circuit 202 board mounted on a thick metal baseplate 204, protected at the front by a coupler face bar 206, which may have a thickness greater than that of the baseplate 204. The coupler face bar 206 may transfer an end-on impact directly to the baseplate 204, thus protecting the circuit board. The circuit board mounting system, not all of which is shown, may be designed to reduce the effects of any forces, including, for example, vibration forces, that would otherwise produce relative movement. The magnetic windings that couple the signal to the next car may be embedded in the single printed circuit board 202. Auxiliary printed circuit boards, corresponding to the circular winding boards used in the existing design, may not be necessary. Apertures or windows in the coupler face bar may permit entry and egress of the magnetic field. An end plate 208 may be mounted on the rear of the bridge unit 100. FIGS. 6 and 7 provide additional assembly details.

[0059] FIGS. 3 and 4 show an exemplary mounting arrangements 300, 400 of the bridge unit 100 as part of the top electrical coupler head 302. FIG. 3 illustrates a single-network installation 300 and FIG. 4 a dual network setup 400 incorporating two bridge units 100. FIGS. 3 and 4 also show a typical arrangement of normal contact pins 302 coexisting with the bridge units. As shown in FIGS. 3 and 4, the bridge unit 100 may be smaller than the normal contact array 302. Environmental sealing of the bridge unit is provided by the mechanical package design, without any interior potting

Magnetics Design

[0060] FIGS. 8 and 9 show two views 800, 900 of the magnetics assembly, which uses an essentially common design for the sender 108 and receiver 106 magnetics. The signal winding in each case comprises two rectangular spiral coils 802, 804 distributed symmetrically about the center plane of the circuit board. These coils are formed by printed circuit traces 806 on

internal board layers. The two spiral coils are counterwound so that their individual flux contributions reinforce one another when the coils are joined in series at their common center. The coil on one board layer spirals inward when traced counterclockwise from the top, while the other spirals outward. A resistor 808 placed in series with the coils at their joining point provides termination for the parasitic nonuniform transmission line formed by the traces. This feature, also present in the existing Etherpin design, suppresses the distributed resonance that would otherwise distort the data waveform. The ferrite polepieces 812 that collect and guide the flux lines may be molded or cemented into each of two plastic covers that seal the magnetic coupling windows in the coupler face bar. [0061] FIGS. 8-10 also show top and bottom copper planes 810 blocking the flux generated by the portion of the winding that lies inboard of the ferrite polepieces. This blocking effect is mediated by the action of eddy currents that circulate within the planes. These currents always generate flux in opposition to that produced by the underlying conductors. As long as the blocking plane does not form a continuous loop about the area enclosed by the full winding, the blocking plane does not impair coupling of the sender to the receiver.

[0062] The blocking plane does, however, reduce the leakage flux of the transformer constituted by the sender winding on the first car and the receiver winding on the second car, thus improving the coupling coefficient across the link while reducing stray coupling to other circuits on the board. The blocking planes extend far enough beyond the edges of the windings to create a local return path for the eddy currents. Otherwise, the blocking effect would be only partially effective

Bridge Unit Circuit Design

[0063] A block diagram 500 of the bridge unit 100 is shown in FIG. 5. A switch chip 502 provides a standardized interface to the external LAN 504, and a digital signal processor (DSP) 506 assists in the performance of most of the low-frequency analog processing functions, for example, AGC, adaptive equalization, and auxiliary communications channel. The DSP also provides a standard serial port for diagnostic data collection, supervisory control, and initial configuration download. This port, hereafter referred to as the supervisory serial port, is provided with its own connector and cable.

[0064) Power is supplied to the bridge unit by means of Power Over Ethernet (POE), IEEE Standard 802.3af, in which 48v DC power is carried by the same cable that transports the Ethernet data. Self-test is facilitated by loopback windings 508 that permit the Ethernet signal to be shunted inductively from the transmitting half-transformer directly to the receiving half- transformer on the same bridge unit. More detail is provided in the section hereinbelow entitled Self-Diagnosis and Fault Reporting.

[0065] In one embodiment, the bridge unit is equipped with two LAN port connections, only one of which is enabled at a given time. This feature may be necessary in a dual network system if space constraints force the two bridge units on each car end to be mounted side-by- side, in contravention of the vertical stacking arrangement shown in FIG. 4. A given bridge unit may carry traffic for one network or the other, depending on the relative orientation of the coupled cars. The two network connectors on each bridge unit, each associated with one of two mutually exclusive virtual LAN (VLAN) configurations of the switch chip, may allow traffic to be steered onto the correct external pathway. Selection of the data path (e.g., Portl to Port2 or Portl to Port3) may be accomplished by the DSP, which outputs logic levels to the VLAN configuration pins on the switch chip.

[0066] The following sections describe in detail the various functions incorporated into the bridge unit. FIG. 11 shows, by means of a data flow diagram, how the DSP firmware may be organized. Automatic Gain Control

[0067] The amplitude of the Ethernet receive signal conveyed by the magnetics may vary over a wide range determined by the spacing and alignment of the bridge units on the two coupled vehicles. The signal amplitude at receive terminals of the switch chip, on the other hand, may be maintained within narrower specifications, with, for example, an AGC loop, which may be implemented in hardware and/or software. The hardware portion may include two elements: a peak detector, connected to an analog-to-digital converter input (ADCl) on the DSP, and a voltage-controlled variable-gain amplifier (VCA) with its control pin wired to a pulse-width modulated (PWM) output (labeled as PWMl in FIG. 5). In the DSP software, the digitized peak detector output may be compared with a reference value, and the gain of the VCA may be adjusted to maintain the detector output at the desired level. FIG. 1 1 shows this

control mechanism as a proportional-plus-integral-mode controller operating by means of numerical calculation in real time.

[0068] The AGC loop may also remove a low-frequency amplitude modulation envelope purposely impressed upon the Ethernet signal to provide the auxiliary communications channel. The AGC loop may be responsive enough to change the VCA control voltage at a rate sufficient to counter these deliberately introduced amplitude variations. In the course of stripping the envelope from the Ethernet signal, the PI controller may recover the envelope for further processing within the DSP.

Auxiliary Communications Channel [0069] The bridge unit may include an auxiliary communications channel for the exchange of operational and diagnostic information between the bridge units on two coupled vehicles. The auxiliary communications channel may be a low-data-rate serial link implemented with an AM envelope impressed on the Ethernet waveform. The sine and cosine phases of the subcarrier may be separately modulated to achieve a combined throughput of, for example, two bits per cycle, yielding a data rate of 640 bits/sec for an exemplary subcarrier frequency of 320 Hz.

This widely used technique is known as quadrature amplitude modulation (QAM). A software modem may be provided within the DSP to encode and decode the transmitted information.

[0070] The auxiliary communications channel may be used in a variety of features. For example, the auxiliary communications channel may be used for exchange of spectral analysis data for adaptive equalization (for more detail, see the section hereinbelow entitled Adaptive Equalization) or exchange of diagnostic information between linked bridge units. In addition, the auxiliary communications channel may be used for transparent linking of the supervisory serial ports on the bridge units in two different cars, thus forming an auxiliary serial trainline. Finally, the auxiliary communications channel may be used for support of consist enumeration and/or for selection of the active network port in another exemplary embodiment of a dual

LAN system, in which the two bridge units that are mounted on an end of each vehicle cannot be vertically stacked.

[0071] Regarding the use of the auxiliary communications channel for support of consist enumeration, when a new car is added to a consist — a train of two or more cars — the supervisory intelligence in each car may identify the car number assignment and order of the

joined vehicles. This task may be accomplished through the Ethernet link, but the process may become quite complex. A direct readout of the ID number belonging to a newly added car, made available to external equipment through the auxiliary communications channel and supervisory serial port, may simplify the identification process. [0072] The term "media access control address" refers to a hard-wired Ethernet address that identifies a piece of equipment to the network. In contrast, an "Internet Protocol address" is assigned to the equipment by a server or set manually through operator action. When a train is put together, the media access control (MAC) addresses of the various pieces of equipment on all the cars may be acquired by the network central intelligence (i.e., the network controller) in each car before train-wide communication can take place over the network. The serial port on the bridge unit, together with the auxiliary communications channel operating between linked cars, may provide the network controller with the car number of a newly added car by a means that does not involve the exchange of actual Ethernet frames over the network. If the MAC addresses of the equipment in the car are keyed to the car number by a known formula, the network controller may determine the MAC address of every piece of Ethernet equipment in the newly added car before any attempt is made to communicate with such equipment over the Ethernet. In an alternative embodiment, equipment identification is done over the network, using only Ethernet traffic, by means of a process that starts with the transmission of broadcast frames. [0073] In an exemplary embodiment of the consist enumeration process, the MAC addresses of the Ethernet equipment in a new car are keyed to the vehicle ID number in a uniform way, so that all the MAC addresses are made available in a single step. The car number may be programmed into the nonvolatile memory of the bridge unit at the time the equipment is installed on the car. The supervisory serial port may provide the means for downloading the car number and other semi-permanent information.

[0074| Selection of the active network port refers to the exchange of car end and vehicle ID information over the auxiliary communications channel in a dual-network system wherein the two bridge units on each car may be arranged side-by-side. In such a system, there may be uncertainty as to which of the two independent networks ends up being carried by a given bridge unit, the association depending on the orientation of the coupled cars. The identifying information mentioned above may be exchanged between link partners as an aid in the routing

of network traffic to the proper LAN port, two of which are provided on each bridge unit. This coordination may insure the continuity of each of the two networks across the coupler. The side-by-side arrangement of bridge units in a dual-network setting may increase the number of cables penetrating the car body. [0075] In FIG. 5, the AM subcarrier supporting the auxiliary communications channel may be impressed on the Ethernet waveform by means of a voltage controlled amplifier (VCA) whose gain control pin is wired to a PWM output (called out as PWM3 in the diagram) on the DSP. The PWM hardware serves as a digital-to-analog converter for the AM envelope, which may include a low-frequency sine wave already carrying a QAM data imprint. The sine wave generation and QAM processes may be performed in real time by software.

[0076] The AGC loop may strip the AM envelope from the Ethernet signal presented to a switch chip receiver but may preserve the modulation waveform for further processing within the DSP. The switch chip receiver may be the receive portion of an internal Ethernet PHY that is part of the switch chip. The switch chip receiver may be different from the receiver that resides in a previous Etherpin application, and may be different from the receiver in the integrated bridge unit of the invention. The QAM-imprinted data may be recovered by a demodulation process incorporating a software phase-locked loop

Adaptive Equalization

[0077] The baseband Ethernet signal may be coupled from one vehicle to another by magnetic induction. The transmit winding on the first car and the receive winding on the second car may form a broadband transformer. Despite the presence of the ferrite polepieces (see, e.g., FIGS. 7-9), the magnetic circuit resides largely in air. As a consequence, the overall frequency response of the transformer is distinctly high-pass, producing a large amount of droop in the data pulses. Restoration of the original Ethernet waveform may require equalization, which may be accomplished by means of a low-frequency boost filter on the receive side of the bridge unit. In the existing Etherpin designs, the equalizing filter uses fixed component values, and successful waveform reconstruction relies on the accuracy of the transformer parameters that govern frequency response. Improved performance in the form of a reduced bit error rate may be obtained, however, by adding some form of dynamic tuning, so

that the equalizing filter always tracks the characteristics of the transformer, even when those parameters vary from bridge unit to bridge unit or over time.

[0078] The bridge unit may include a novel magnetics design. The magnetics design in accordance with the invention may exhibit a greater need for adaptive equalization than do known magnetic designs, mainly because of a higher coupling coefficient of the magnetics design in accordance with the invention. The advantages gained by coupling more of the magnetic field to the opposing bridge unit may be an improved signal-to-noise ratio and a reduced gain requirement in the receiver. The penalty paid may be an increased variability of winding inductance with air gap, resulting in wider variations in frequency response. On a train, the distance between sending and receiving coils (the air gap) may be affected by mechanical tolerances in the coupler head and, in some cases, varies slightly with pulling force. Therefore, the bridge unit may include a form of adaptive equalization.

[0079] The receive equalizing filter shown in FIG. 5 may include a varactor diode tuning mechanism controlled by means of a voltage output from a PWM port on the DSP chip (called out as PWM2 in the diagram). This control voltage may be calculated by a software servo loop that compares a local frequency analysis of the received Ethernet waveform with a corresponding analysis of the transmit signal at the other end of the link. The information exchange required by this comparison may take place over the auxiliary communications channel. [0080] The receive analysis filter in the first of two coupled bridge units may feed an RMS detector connected to an analog-to-digital converter input (ADC2) on the DSP. The RMS detector may measure the receive signal energy across each of two frequency bands selected by the filter. The filter may be programmed through a normal output port pin configured to pass either a fairly narrow band of low frequencies (e.g., 50 kHz to 1 MHz) to the detector, or a wider band (e.g., 50 kHz to 10 MHz). The wideband measurement may provide a reference level to permit ratiometric compensation for filter component tolerances as well as for deviations in the detector response. The primary analysis result may be the relative low- frequency energy content of the signal, which governs droop and baseline wander in the waveform presented to the switch chip. Cable equalization, which is primarily a high- frequency function, may be left to the PHY of the switch chip.

[0081] The transmit analysis filter in the first bridge unit may operate in a similar way to measure the relative low-frequency content of the Ethernet signal transmitted to the second bridge unit of the coupled pair. The analysis results may be conveyed to the second bridge unit by means of the auxiliary communications channel. The second bridge unit may communicate its own transmit analysis to the first bridge unit in the same way. Comparison of the two low-frequency energy measurements (local Rx versus distant Tx) may tell the DSP which way (and by how much) to adjust the control voltage that tunes the receive equalization filter. A software servo loop in each bridge unit may continuously run this adjustment algorithm to minimize the difference between the two measurements. [0082] There may be a delay of a few milliseconds associated with the auxiliary communications channel. This delay may be deterministic, and so its effects may be mitigated by setting up the measurement and communication sequence in accordance with the following steps. In step (a), the bridge unit in Car 1 may initiate a measurement of its transmit signal. In step (b), the bridge unit in Car 1 may set an internal timer to a value T| and send a short "heads up" packet to the bridge unit in Car2. In step (c), upon receipt of the "heads up" packet, the bridge unit in Car2 may set its own internal timer to a value of Ti -Tc, where Tc represents the known communications delay. In step (d), when the timer in Car 1 expires, the bridge unit in Car 1 may begin measuring its transmit signal, accumulating these numbers into a single sum- of-squares total. In step (e), when the timer in Car2 expires, the bridge unit in Car2 may begins measuring its receive signal, accumulating these numbers into a single sum-of-squares total. In step (f), both bridge units may accumulate the same number of readings. This number may be large, because the spectral energy in the low-frequency band may vary considerably over short time intervals. Because of the timer synchronization method just described, the two measurement processes may run essentially in parallel, operating on the same Ethernet symbol sequence at two different locations. In step (g), the bridge unit in Car 1 may transmit its measurement results to the bridge unit in Car2. In step (h), the bridge unit in Car2 may compares its own measured energy total to the number reported from Car 1 and updates the control output to its equalization filter.

[0083] A similar sequence may be driven concurrently from the Car2 side, thereby allowing the bridge unit in Car 1 to update its own receive equalization. The equalization adjustment process may not be particularly fast. A reasonable sampling rate is, for example, one sample every 500 ms.

Self-Diagnosis and Fault Reporting

[0084] The loopback connection shown in FIG. 5 facilitates built-in test of the entire signal path through the magnetics, amplifiers, and filters. When the loopback amplifier is enabled, inductive coupling between the transmit and receive windings on the same bridge unit takes place by way of the additional test windings embedded in the printed circuit board (see FIG. 9).

[0085] Stimulation of the analog receive path by the means described above may permit the DSP to test the hardware components of the AGC loop, along with the frequency analysis filters and RMS detectors used for adaptive equalization. The LED driver pins of the switch chip may be read by the DSP to confirm detection of the loopback signal by the switch chip. The auxiliary communications channel may be used to share diagnostic results between cars when one car becomes unreachable over the network (due, for example, to a high bit error rate on the Ethernet link) but may remain accessible over the low-speed channel.

[0086] When a link partner (e.g., a second bridge unit) is present and the auxiliary communications channel is operational, there may be no need for a loopback test. The successful operation of the auxiliary communications channel may be sufficient proof of the integrity of the transmit signal path and of the receive signal path on both ends of the link. However, a "gray" fault, such as a lower-than-normal receive signal level in one of the coupled bridge units, may still be noted by diagnostic routines in the DSP and may be shared with the link partner so that the fault and its location is reported in both coupled vehicles. This sharing of diagnostic data between cars may be successful even if the Ethernet link has a very high bit error rate. It may be necessary only that the Ethernet waveform be capable of supporting the amplitude modulation envelope that carries the auxiliary communications channel.

[0087] When no link partner is detected, the loopback circuit may be used to enable excitation of a receive section of the bridge unit by the local transmit signal. The functions of

the AGC, equalization, and auxiliary communications channel may then be checked out even on an isolated rail vehicle. Execution of the self-test may be initiated, and the results collected, through the supervisory serial port. This port may be connected to a portable test unit for field service or to an on-board data collection system, which is the MDS, for continuous fault monitoring.

Characteristics of the Fast Ethernet Line Signal

[0088] The Fast Ethernet standard, represented by its most prevalent implementation, 100Base-TX, encodes data for transmission in a three-level form known as MLT-3 (Multilevel Transition-3). FIG. 12 illustrates an idealized waveform 1200 showing the general concept. Under this encoding scheme, a logic T produces a transition on the line, while a logic '0' maintains the previous state. There are three voltage states: +1V (1202), OV (1204), and -IV (1206). Referring to FIG. 13, a series of transitions traverses the state list 1300 in the following order: OV (1302), +1V (1304), OV (1306), and -IV (1308). A continuous series of T bits thus produces a cyclic pattern of states with a fundamental frequency of 31.25 MHz. A continuous series of '0' bits, without alteration, would maintain the line at a constant voltage level for an indefinite time. In order to avoid this situation, a pre-encoding scheme called 4B5B is used to transform the raw data by adding a fifth bit to each nibble. The extra bit is assigned a value that insures the presence at least a single logic '1' in each five-bit group. Hence, a continuous stream of zeroes at the raw data level will insure a minimum line frequency of 8 MHz, as shown by the waveform 1400 in FIG. 14.

[0089] Continuous transitions facilitate magnetic coupling of the signal as well as clock synchronization in the PHY receiver. The rise time of the MLT-3 signal is critical and must be maintained within the range 3-5 ns at the transmitter. The actual data rate on the line is 125 Mb/sec, rather than 100 Mb/sec, due to the insertion of the fifth bit. [0090] The clean, crisp waveforms shown in FIGS. 12 and 14 may be observed only at the transmitter or at the end of a very short cable. Moderately long cables produce rounding of the pulse edges, and the longest ones destroy any resemblance to the images shown. Commercial Ethernet PHY circuits (incorporated into every hub, switch, router, or network interface card) restore the original transmitted waveform by means of adaptive equalization, automatic gain control, and baseline wander compensation.

[0091] The standard cables used for transmission of Fast Ethernet are categorized as CAT-5 or CAT-5E and consist of multiple twisted pairs with a characteristic impedance of 100 Ohms. A single pair is used for data transmission in one direction and a second pair for transmission in the opposite direction. Two additional pairs are also present, but these usually remain unused.

The Standard Ethernet Transformer

[0092] For the sake of isolation, Ethernet is usually coupled onto the cable by means of wideband pulse transformers. These transformers consist of a few turns (typically 10) of small-gauge solid magnet wire on a pair of tiny ferrite cores, as illustrated by the transformer 1500 in FIG. 15. One core is used for the transmit signal and the other for the receive signal.

[0093] Typically, 2-8 such transformers are bundled together in a plastic integrated-circuit package. Additional components such as common-mode chokes are often included. Many RJ- 45 connectors that mate with the CAT-5 cables come equipped with integral magnetics.

[0094] The most significant property of these transformers is their extremely tight coupling (i.e., high ratio of mutual inductance to leakage inductance). The following specifications are fairly standard:

(0095J The effective leakage inductance at very high frequencies must be even lower than the data sheet value, or else these transformers would not perform as well as they do. Without extremely tight coupling, a transformer cannot achieve faithful reproduction of a wideband pulse train. Magnetic Coupling Across an Air Gap

[0096] There are circumstances under which it is necessary to couple a wideband pulse stream such as Fast Ethernet across an air gap between two moveable pieces of equipment. In these cases, the tight coupling provided by a standard Ethernet transformer is not available. It is therefore necessary to develop means of correcting for the pulse distortion that is inherent in a high-leakage transformer design. Two kinds of distortion must be dealt with: frequency distortion and resonance. The first can be handled by means of fairly straightforward equalization techniques, which will be considered later. The second form, considered below, may be more difficult to remedy. FIG. 16 illustrates a graph 1600 of an output signal 1602 featuring both resonance distortion 1604 and frequency distortion 1606. [0097] The adaptive capabilities of a standard Ethernet PHY circuit may not be relied upon to correct for these forms of pulse distortion because the PHY may be designed to handle the kinds of distortion produced by a specific type of cable. Commercial PHY chips contain a frequency-dependent reference model that mimics the attenuation per unit length of CAT-5 or CAT-5E. Frequency distortion that does not fit this profile may be corrected before the signal is presented to the chip.

Fundamentals of Distributed Resonance

[0098] Regarding transformer resonance, each winding may be viewed as a long, nonuniform transmission line. Lumped resonance models do not accurately describe the behavior of a loosely-coupled transformer when it is excited by a narrow pulse. In fact, instead of in terms of electron drift, a conducted pulse may be visualized as a charge density wave, with accompanying EM fields, in the electron gas. The concept of charge density waves is usually invoked in the study of certain exotic phenomena in semiconductors, but it may be equally useful for the purpose of analyzing the behavior of ordinary transmission lines. Ultimately, a simple way of calculating a critical circuit value that would be very difficult to determine by other means is gained.

[0099] A narrow pulse applied to one end of a pair of wires reaches the other end in far less time than it takes an electron to drift the same distance. Typical electron drift velocities are on the order of 1 mm/s, while the speed of signal transmission along a cable often exceeds 2/3 of the speed of light. When we suddenly push a bundle of charge 1702 into one end of a wire, an equivalent bundle 1704 appears at the other end almost instantly, as illustrated by the system 1700 of FIG. 17. There has been a net transfer of charge, but the charge packet delivered to the load 1706 is not the same charge packet that was injected into the wire by the source. As an analogy, a similar phenomenon occurs when a piston is driven suddenly into one end of a pipe, as illustrated by FIG. 18. A slug of air 1804 is expelled from the far end after a delay determined not by the velocity of the piston 1802, but by the speed of sound in air. What propagates down the pipe is a pressure pulse 1806, which may also be termed a "fluid density wave." The main difference between the hydrodynamic and electrodynamic examples is the nature of the force that propagates the effect in each case. In air, the force is purely mechanical, described by the laws of gas dynamics. In the wire, the force that acts on the charge is electromagnetic, described by Maxwell's Equations.

[0100] FIG. 17 illustrates that the leading edge of the negative charge density wavelet moves forward by drawing in negative charge from the electron gas reservoir 1708 that fills the conductor. The trailing edge propagates in the same direction by returning electrons to the reservoir. The local free-charge currents associated with this process are matched by displacement currents in the dielectric medium between the conductors. This continuity of charge flow across the boundary is a three-dimensional expression of Kirchoff s nodal circuit law. The total current field associated with the wavelet pair forms a nanoscale vortex that moves along the inter-conductor channel like a miniscule tornado, incorporating free charge from the metal and bound charge from the dielectric. [0101] Three different conditions of uniform transmission lines will now be considered: a shorted line 1900, as illustrated by FlG. 19, a matched line 2000, as illustrated by FIG. 20, and a mismatched line 2100, as illustrated by FIG. 21. An ideal current source is used to drive the line with a short, square pulse.

[0102] In the shorted case, referring to FIG. 19, two charge density wavelets 1902, 1904 propagate down the line simultaneously toward the shunt 1906: a positive wavelet 1902 emitted by the plus terminal of the generator 1908, and a negative wavelet 1904 emitted by the

minus terminal. Both polarities of the wavelet may be present, because any source driving a closed circuit experiences equal and opposite current flows in its plus and minus leads at all times. At the positive terminal, a local rarefaction of the electron gas takes place, while at the negative terminal, there is a local compression. Both disturbances propagate toward the shunt at a speed determined solely by the dielectric permittivity and magnetic permeability of the medium surrounding the conductors. The charge wavelets travel near the surface of the conductor due to electrostatic forces that exclude unbalanced charges from the interior. In this respect, the charge density pulse may more closely resembles a water wave traveling along the surface of a pond than it does a pressure wave propagating inside a pipe. [0103] When the two charge density wavelets 1902, 1904 reach the shunt 1906, they pass through one another and loop back toward the source, each taking the leg of the transmission line opposite to the one on which it was initially transmitted. The two wavelets 1902, 1904 do not annihilate one another, because there may be no resistance in the shunt to dissipate their energy. Even though the total charge density at the moment of coincidence goes to zero, the charge packets reform and continue on their way because of the energy stored in the magnetic field. Oppositely phased water waves have a similar way of passing through one another without losing energy, due to the undercurrents associated with each crest and trough.

[0104] These vortices preserve the wave energy in kinetic form when cancellation of the vertical displacements drives the potential energy to zero. Long before the charge wavelets arrive back at the source terminals, the current source has turned off, forming a passive open circuit. Each wavelet reflects back from this barrier like a water wave hitting a sea wall. The two of them once again propagate toward the shunt, where they repeat their intermingling passage and return to the source along their original paths. This process of reflection and interpenetration continues forever, since there are no losses in the system. This description recapitulates what is already known to happen in an ideal shorted line while providing additional insight into the physical processes involved.

[0105] The next case to consider is the matched line. Here, referring to FlG. 20, the initial density wavelets 1902, 1904 propagate toward the end of the line as before and, upon entering the terminal region, once again interpenetrate. This time, however, there is energy dissipation due to the nonzero resistance of the load 2002. If the resistance value is properly selected, the

Poynting energy of each wavelet will be absorbed and converted to heat by the time the wavelet has propagated all the way through the load from one conductor to the other.

[0106] The final case is the mismatched line, illustrated by FIG. 21. Here the wavelets interpenetrate within 1902, 1904 the resistive load 2002 and heat is once again released. This time, however, not all the Poynting energy may be dissipated. If the load resistance is too high, some of the wave energy will flow back into the electric field as potential between the conductors. If the resistance is too low, some of it will remain in the magnetic field as kinetic energy of charge transport. In either case, a pair of diminished wavelets will return to the source through the usual path exchange process and then reflect back toward the load. This sequence of decaying echoes will continue until all perceptible energy has been dissipated.

The subject of transmission line matching, it turns out, is central to the methodology of dealing with resonant distortion in pulse transformers. In later sections, we will investigate the matter quantitatively and extend our analysis to nonuniform lines.

Windings as Nonuniform Transmission Lines [0107] FIG. 22 illustrates an system 2200 featuring a single-layer solenoid 2202. When excited by a pulsed current source, the winding experiences the arrival of charge density wavelets on both of its leads, one lead receiving a positive charge injection 2204 and the other a negative one 2206. These wavelets follow oppositely-oriented helical paths, converging at the midpoint 2206 of the winding and passing through one other as in the earlier case of the shorted transmission line. Upon emerging from the positive lead of the solenoid and reaching the now-inactivated positive terminal of the source, the negative charge bubble reflects back. The same thing happens to the positive charge bubble upon reaching the negative source terminal. The charge wavelets continue to bounce back and forth through the solenoid winding, always crossing at the midpoint. This description elucidates the mechanism behind the ringing of the coil in response to the sudden excitation of its distributed capacitance and inductance. The transmission line model still applies, but the geometry is completely nonuniform. In a uniform line, the injected positive and negative charge wavelets travel together and the distance between them remains constant. In a nonuniform line, this distance continually varies. The midpoint of the winding corresponds to the terminal shunt in the shorted line model.

[0108] The two-layer printed spiral winding 2300 illustrated in FIG. 23 affords another example of resonance mediated by charge wavelets. In this case, the center printed-circuit via 2304 connecting the top and bottom circuit layers represents the midpoint of the winding 2302 and corresponds to the terminal shunt of the equivalent transmission line. [0109] Recognition of the distributed resonance of the winding as a transmission line phenomenon leads naturally to the idea that the ringing of the equivalent line (and hence of the actual winding) might be suppressed by the addition of a suitable terminating resistor 2402 at the winding midpoint (as shown by the winding 2400 in FIG. 24). This measure has indeed been found quite effective in reducing the resonance distortion of a loosely-coupled Ethernet pulse transformer. The difficult trick, however, is to calculate the proper resistor value. That development will occupy us for the next two Sections.

Analysis of a Uniform Transmission Line

[0110] The universal matching criterion for all lines, both uniform and nonuniform, is Power Dissipated = Power Propagated. The left-hand side of the equation represents the power dissipated in the load and the right-hand side stands for the Poynting flow from source to load through the line. In the case of a uniform line, the situation 2500 is depicted by FIG. 25, in which L ₀ stands for the inductance per unit length, C ₀ is the capacitance per unit length of the line, and v _p is the velocity of propagation. The line is driven by a stepped current source with strength I. Prior to time t = 0, the current delivered by the source is 0; afterward, it remains constant at the value I. Standard transmission line analysis yields that v _p = .

Generalized Transmission Line Analysis

[0111] FIG. 26 depicts a transmission line of arbitrary geometry driven by an ideal current source delivering a step waveform of amplitude I. The steady state will be modeled by a continuous train of charge density wavelets traversing the line from end to end. Let t _p stand for the propagation time from source to load and δQ for the total charge transported by the wavelet train during a time interval δt = t _p . Standard transmission line equations yield R ₁ = ^p /-, , the critical value of a non-reflecting load.

[0112] The energy partition theorem — L ₀ I ₂ = -C ₀ V ² for a uniform transmission line may be

generalized to a universal form. First, the energy stored in the magnetic field of a line or winding may be separated into two components. Within any small segment of the line, the first component of the local magnetic energy density is the one contributing to the Poynting flow along the conductor. The Poynting vector, according to standard EM theory, is defined by the formula S = — E x B , where S points in the direction of wave propagation and is M expressed in units of energy per unit area per unit time. The electric field E and the magnetic field B appearing in the equation are the fields generated by the source charges producing the wave. The magnetic permeability μ is a property of the medium surrounding the conductor in which these source charges reside. The Poynting vector calculated by applying the above equation to the near fields of a charge density wavelet identifies the local energy flux along the conductor.

[0113] The second component of the magnetic energy density in the vicinity of a given line element is associated with fields propagating at right angles to the local conductor. Such transverse energy flows do not exist in nonradiating uniform lines, but in coiled or convoluted geometries, there is cross-coupling between every pair of line elements. This coupling does not contribute to the Poynting flow along the conductor, but it does induce a voltage in each local element that contributes to the total EMF measured at the source terminals. In a spiral- wound transformer, the induced voltage in each turn is the sum of contributions from all the turns, but the energy flow along the conductor at any point is due solely to the fields carried by the local charge density wavelet. Therefore, the total inductance of a terminated winding (or of an arbitrary system of conductors) may be separated into two components: an intrinsic self- inductance that participates in the flow of energy toward the termination resistor (load), and a mutual self-inductance that stores additional energy which is not communicated to the resistor. In an isolated, shielded winding, one may regard the field energy stored in the first inductance component as mobile or propagating and the second energy component as trapped. If the winding termination resistance is selected according to the above equations, the back-EMF produced by the flowing energy will be resistively phased, while the EMF due to the trapped energy will be in quadrature with (i.e., functionally orthogonal to) the source current. Under DC excitation, this statement is obviously true, because the terminal voltage of a coil

terminated resistively at its midpoint will be equal to IR _t , regardless of the inductive field energy stored in the coil. When the wound transmission line is part of a transformer, a portion of the "trapped" energy component is actually radiated to the secondary, but this propagation takes place in a direction transverse to the primary current flow.

t ² [0114] We may now determine that the intrinsic self-inductance, L ₁ , is given by L ₁ = — . The

propagation time t _p is given byt _p = -^μ _r ε _r — , where 1] is the conductor length, c is the speed of light in a vacuum, and μ _r and ε _r are the relative permeability and permittivity, respectively, of the medium surrounding the line. The aggregate capacitance C can easily be measured, or, alternatively, calculated from the conductor geometry and the dielectric constant of the medium.

Energy Flows in a Bipianar Spiral-Wound Transformer

[0115] FIG. 27 shows an idealized version of the primary winding structure 2700 used in the invention, including the conductive shield and ferrite field-stop. The main point of this illustration is its portrayal of the energy flows within the transformer. In the previous section, it was noted that the energy stored in an isolated, terminated primary winding may be divided into two components, one propagating and the other trapped. The trapped energy, as FIG. 27 shows, takes the form of a radial standing wave bouncing back and forth between the shield and the center printed circuit via. This energy pool accounts for most of the terminal inductance of the coil, since it arises from the inter-turn coupling that makes the total inductance proportional to turns-squared. When a loaded secondary winding is present, some of the trapped energy is coupled to the secondary circuit and dissipated. Three energy flows may be present in the primary winding: (1) the tangential flow of energy along the spiral conductor to the terminating resistor(s); (2) the radial (oscillatory) flow of trapped energy within the cavity formed by the two primary winding layers, the center via, and the shield; and (3) the vertical flow of coupled energy to the secondary winding (located above the primary but not shown in FIG. 27).

[0116] Three modes of resonance are possible in this configuration. In the current embodiment of the invention, a mismatch between the winding and the termination resistance — associated with item (1) above — may cause quarter- wave ringing at about 130

MHz. This effect may be reduced by the insertion of the proper termination resistor at the winding midpoint. The second resonance mode is associated with item (2) and may be harder to reduce. The current design, however, places this frequency at around 4 GHz (quarter- wave). This value is outside the system bandwidth and therefore may pose no threat to signal integrity. The third form of resonance accompanies the flow of energy to the secondary. Not all of the vertically propagating energy is absorbed and dissipated; some is reflected off the secondary ferrite (not shown). The two ferrites form an air-filled resonant cavity that rings at about 15 GHz, well outside the operating bandwidth. Moreover, this mode is heavily damped due to the lossiness of the ferrite material at such frequencies. [0117] Also shown in FIG. 27 are vector diagrams revealing the constituent field components associated with each energy flow. The corresponding Poynting relationships are expressed by

-E _z x B _r = S ₀ (tangential flow), — E, x B ₀ = S _r (radial flow), and — E ₁ x B _r - S ₂ (vertical

Mo Mo Mo flow). These relationships represent the actual energy flows described above only to the extent that the spiral winding may be approximated by a set of concentric circles. For a coarse spiral, absolute rigor would demand the introduction of a special coordinate system in which ii ~ is replaced by a unit vector tangent to the exact curve of the winding.

[0118] In the equation for tangential energy flow along the conductor, the z-component of the electric field at a given point on the surface of the metal is proportional to the surface charge density, an observation which identifies the vector ε _z as belonging to the local field system of a charge wavelet. Likewise, the r-component of the magnetic field at the same boundary point is due almost entirely to charge transport by the underlying wavelet. Hence, the Poynting component So may be taken as an excellent representation of the power flowing toward the termination resistor RI.

[0119] Regarding the equation for radial flow, the magnetic field component parallel to the conductor, B ₀ , is quite weak and is due exclusively to the vertical current flowing in the center via connecting the two winding layers. Furthermore, it interacts energetically only with the small vertical electric field E _z ' created by the inductive voltage drop along that same path (the larger portion of E _z being due to charge wavelet activity in the spiral traces). The resulting standing-wave resonance must therefore be only weakly excited. That inference, coupled with

the earlier statement about the very high frequency of this mode, lays to rest any remaining concerns about ringing due to radial energy flow.

[0120] Regarding the equation for vertical flow, if the transformer is loosely coupled, the energy transported vertically to the secondary will generally be less than the portion flowing along the primary conductor. However, when the two windings are in very close proximity, loading of the secondary can further improve the damping of the transformer. Excessive loading, however, may degrade the signal bandwidth.

The Purpose of the Ferrite

[0121] The purpose of the ferrite is to straighten the flux lines, increase the inductance by a slight margin, and prevent stray magnetic coupling to nearby circuits within the sender or receiver housing. The exact properties of the ferrite are of minor importance, since the air gap accounts for about 80% of the magnetic circuit reluctance. Nevertheless, a good pulse- transformer ferrite should be chosen. The material used in the current embodiment of the invention is a nickel-zinc ferrite with an initial permeability of approximately 1300 at 50 kHz, 100 at 100 MHz, and 10 at 1 GHz. The Q (an inverse measure of lossiness) may range from, for example, 23 at 50 kHz to 0.32 at 100 MHz.

Immunity to Contaminating Media

[0122] The near-field coupling system used in this invention demonstrates excellent immunity to contamination by snow, ice, water, and iron-loaded dust. Even a concentrated saltwater solution has little effect on signal throughput. This robustness is believed to be superior to that of a true wireless solution based on a high-frequency RF carrier. The following reasons may support this claim. First, a wireless system capable of transporting 100-megabit Ethernet uses a carrier frequency comparable to that of 802.1 Ig, which operates at 2.4 GHz. At this frequency, the quarter-wave distance in air is 1.23" and less than 0.2" in watery snow. Therefore, far-field models must be used even with very close antenna spacing. Second, in a far-field link, the local magnetic field is supported almost entirely by displacement currents in the transmission medium, making the channel performance very sensitive to the dielectric and conductive properties of the medium. In a near-field link, the local magnetic field is supported mainly by charge wavelets in nearby metallic conductors, relegating the magnitude of the displacement current effect to a distant second order. This

advantage is compounded by the lower frequency of the baseband signal, as compared with that of an RF carrier.

Circuit Model of the Transformer with Distributed Resonance

[0123] FIG. 28 shows an equivalent circuit 2800 of the transformer used in circuit simulations. The principal resonant mode of each winding is modeled to the first order by a uniform transmission line element having a characteristic impedance equal to the termination resistance calculated for the actual winding. The equation used for this purpose is the termination resistance equation. The method of determining the aggregate winding capacitance is modified slightly for the secondary winding, which is a single-layer coil. [0124] In FIG. 28, circuit elements O _p i and O _p2 represent a pair of single-ended transmission lines joined back-to-back to form a single differential line. The end-to end inductance of the differential line is chosen to equal the intrinsic self-inductance of the primary winding as defined by the intrinsic self-inductance equation. The complete terminal inductance measured between terminals pri+ and pri- (which includes the "mutual" self-inductance as defined earlier) is formed by scaling up the end-to-end voltage drop along the differential transmission line. Dependent sources E _p i and E _p2 were introduced for this purpose.

[0125] Next, assume that R _pt i and R _pt2 have been shorted. At moderate frequencies (up to, e.g., 10 MHz), the drop along the transmission line is almost purely inductive, and therefore the series combination of line (O _p i ,O _p2 ) and dependent sources (E _p i, E _p2 ) presents an essentially pure inductance to the winding terminals across this portion of the spectrum.

However, when a sharp current step is applied to the pri+/pri- terminals , a slug of increased charge density propagates along O _p i/O _p2 , developing an end-to-end voltage that increases with time. This voltage tracks the progress of the inductance horizon, marked by the propagating charge wavefront as it moves toward the terminating resistance (Rpti/Rpα). A scaled-up copy of this increasing voltage drop is reflected in EpI and Ep2, accounting for the "trapped" inductive energy in the coil that does not propagate along the spiral conductor but whose buildup accounts for most of the induced terminal voltage. As described above, the "trapped" energy is due to mutual coupling between the turns. It takes the form of standing waves whose forward- and backward-propagating components have Poynting vectors directed at right angles to the spiral PC trace. By modeling the "mutual" self-inductance in this way, rather than

simply adding a pair of fixed inductors in series with the transmission line, we obtain a "trapped" field that develops at the same rate as the charge wavefront in the line. Until the entire spiral has been filled with current, the interturn coupling field cannot reach its full extent. At the moment the conduction wavefront touches the termination point (i.e., winding center), the current distribution in the line has fully developed , and so has the magnetic field containing the trapped energy. This timing is not a bad approximation to the actual situation, although the use of a linear transmission line model does not reproduce all the characteristics of the actual nonuniform line comprised by the winding.

[0126] Once the charge density wave train initiated by the current step has reached the winding midpoint, ringing (as described above) begins due to the shunting of R _pl i and R _pt2 . The ringing waveform is strongly magnified at the primary terminals due to the amplification effect of sources E _p i and E _P 2, which simulate the compounding of voltage by the multiturn winding. If resistors R _pt i and R _pt2 are restored to their normal function, the ringing may be quenched and the transformer model passes a clean signal. The secondary winding model is based on the same theory and need not be discussed in detail. Coupling between primary and secondary is modeled by dependent behavioral current sources B _p i, B _p2 , B _s i, and B _s2 . The explanation is that the voltage induced in the secondary by current flowing in the primary of a transformer may be modeled by placing a dependent current source, scaled to reproduce a certain fraction of the primary current, across the secondary winding inductance, which is paralleled by a load. If the load is a short circuit, all the source current is diverted through that path and the output voltage is zero. If the load is an open circuit, the output voltage will be proportional to the secondary inductance times the rate of change of the primary current. For intermediate values of secondary load resistance, the source current will split between the secondary inductance and the load, producing intermediate values of output voltage. If the inductive reactance of the secondary is large compared with the load resistance, the bulk of the source current will flow through the load, producing an output voltage proportional to the primary current times the load resistance.

[0127] Reverse coupling from the secondary to the primary can be modeled in a similar way. When both forward and reverse coupling sources are included in the model, the complete functionality of the transformer emerges. Zero-volt sources V _p i-V _p3 and V _s i-V _s3 are current-

measuring elements required by the rules of SPICE. The currents through these elements appear as arguments to the behavioral functions defining the outputs of B _p i, B _p2 , B _s i, and B _s2 .

Mid-Frequency Model of the Transformer for Equalization Purposes

[0128] FIG. 29 shows a simplified version of the transformer equivalent circuit, valid across the equalization band (45 kHz-30 MHz). The distributed capacitance of the embedded transmission line, which plays a significant part only at much higher frequencies, has been replaced by a single capacitor. The inductive properties of the line have been incorporated into a fixed inductor that represents the sum of the intrinsic and mutual self-inductances of the winding as defined above. FIG. 30 contains a single-ended equivalent circuit that simplifies mathematical treatment. Finally, FIG. 31 incorporates a shunt RC network across the primary for admittance equalization and a resistive load on the secondary for bandpass shaping, and is used in the below analysis.

Design of the Sender Admittance Equalization Network

[0129] The sender admittance equalization network consists of resistor R _eq and capacitor C _eq in FIG. 31. Its purpose is to adjust the driving point admittance of the transformer primary so that the sending amplifier sees an essentially resistive load across the entire Ethernet bandwidth (45 kHz - 100 MHz).

[0130] Circuit analysis of FIG. 31 shows that the admittance presented by the loaded transformer at the primary terminals (without the equalizing network) is given by

S - S _n

Y P-P ( ^S ) = K PP

(S - S^ )(S - S ₂₎

where

[0131] DC resistances R _p and R _s may be small enough to be safely ignored, and the effect of Cpt has likewise been neglected for the purpose at hand. The factors k _sp and k _ps stand respectively for the forward and reverse current transfer ratios. In FIG. 31, the behavioral functions for dependent sources Bp and B, may be written in terms of these factors , whose values come from the detailed transformer magnetics model:

F ₁ ^ (I(VJ) = k _p J(V _s )) _a nd F _s = (I(V _p )) = k _ψ I(V _p )) .

[0132] Continuing with the analysis, we note that the admittance of the equalization network Req, C _cq may be written in the form s 1 1 γ«, ( ^s ) = ^κ «, » ^{where κ} «, = -T- ^{and s} «, =

S - S R «/ K ₁ C,, The total admittance seen by the driving amplifier is therefore given by

S - S ₀

Y _m is) = Y _pp (S) + Y _eq (S) = K _pp ^» + K _eq

(s - s _λ )(s - s ₂ ) s - s _eq

Our goal is to choose K _eq and S _eq so as to make Y _peq (s) as nearly resistive as possible.

[0133] A partial fraction expansion of the first term in the rightmost member of the above equation allows us to write the equalized admittance in the form

Y _ v- ^S o ^{~ S} \ 1 . y ^S 2 ^{~ S} ϋ * . v- ^S

S — S ₁ S - S ₁ S ₂ — S ₁ S — S _eq S — S _eq

in which Si stands for the first corner frequency (typically equivalent to a few megahertz) and s ₂ for the second corner frequency (typically a few hundred megahertz). It is apparent from these relative magnitudes that the middle term will be quite small compared to the first. It therefore makes sense to combine the first and third terms in a way that will minimize the imaginary component of that sum. A good first step in that direction is to set s _eq = s _\ , so that the above equation becomes

[0134] It now becomes obvious that if we set

^K PP S ₀ - S, _ _n

TS ~ 1 '

the first term will reduce to the simple real constant IQ _q , eliminating any imaginary component of the admittance except for the small contribution from the second term. Note that imaginary terms may appear when s is set equal to jω for sinusoidal excitation. Solving for K _eq gives us

K e,q = -K " p.p ^S ° ^S] s _λ (s ₂ - S _x )

All of the frequency parameters S _n are negative, and the order of absolute values is ^ _{0 2} » s The two frequency differences on the right-hand side are therefore both negative, and the sign of K _eq is positive, as required.

[0135] Finally, combining the above equations allows us to solve for the values of the equalizing resistor and capacitor: [0136] FIGS. 32 and 33 show the numerical accuracy of the equalized admittance produced by this solution. The solid curve plots the magnitude or phase of the unequalized transformer admittance Y _pp . The dashed curve portrays the equalized value, with the R _eq C _ςq network added. The corrected magnitude and phase are flat all the way out to 1 GHz.

Analysis of the Transformer Gain as a Function of Frequency [0137] With the admittance equalization network in place, the effect of sender amplifier loading on system frequency response becomes negligible. Therefore, the equalization of the

sender-receiver throughput gain can proceed directly from a refinement of the approximate gain formula

where A _sp stands for the approximate gain from the primary terminals to the secondary terminals. The approximation consists in the omission of lumped capacitance element C _pl . The formula is obtained by a circuit analysis of FIG. 31. Here the constant K _sp is defined by

the formula K _ψ All other symbols in the above equations retain their earlier meanings.

[0138] The above equations can be written in a more compact form by introducing the ancillary symbols

V

1

^S p, \ ~

R _Pl c _PI ^'

R ₁ .

L '

yielding

~ ~ s ~ k

^A v = Ksp — J and ^ _v = --f s _{pl] Spl}

[0139] Wc can now add the capacitance C _pt back into the gain formula by replacing the resistance R _pt wherever it appears with the AC impedance Z _pt which is defined by

- s pi 2

Z P _n i, = ^■ = R P ⁱ

^ pi ^{S S} pι2 S - S pi 2 according to the law for parallel impedances. Thus,

in which the tildes above A _sp and K _sp have been dropped to indicate that these quantities are now exact. A more compact form of the above equation is

[0140] The problem with the above equation is that the insertion of the frequency dependent impedance Z _pl introduces another pole frequency. This extra pole makes subsequent calculations very messy, and so a second-order approximation is sought by repositioning the two main poles so as to match the three-pole response as closely as possible without actually adding an extra pole. The key to this refinement is an approximation to Z _pt that takes the form of a truncated Laurent series,

[0141 ] A typical value of S _pt2 (in cyclic frequency terms) is 200 MHz, and so the approximation is a fairly good one throughout the equalization band (45 kHz - 30 MHz). The final form of the gain equation (after dropping the small cubic term in the denominator) is

where

The overbars indicate an approximation that is still not quite exact. The denominator of above equation may be written in factored form, so that the gain formula becomes

A _p (s) = K ^ — , in which

Receiver Gain Equalization Strategy

[0142] FlG. 34 shows the system gain before and after equalization. This chart is introduced first in order to convey the magnitude of the equalization adjustment that must be made in order for the Ethernet data to be restored to usable form after passing through the sender/receiver transformer coupling system. The solid curve indicates the uncompensated

response of the transformer. The dashed curve shows the final result after equalization. Note that at 50 kHz, the equalizer may provide about 40 db of boost. The low-frequency end is very important, because without sufficient low- frequency response, the baseline wander of the Ethernet waveform becomes too severe for the switching hub PHY receiver to cope with. The equalized curve corresponds very closely to the frequency response of a tightly-coupled, packaged Ethernet transformer.

[0143] At first sight one is tempted to conclude that a single equalization stage with a zero at frequency S i and a pole at a much lower frequency would be sufficient to extend the flat part of the frequency response down to the low end as required. This strategy is unworkable for two reasons. First, too much gain would be required from the single equalization amplifier stage, and, second, the shape of the frequency response (and, more importantly, the corresponding time-domain step response) is too sensitive to small deviations in the equalizer corner frequencies.

[0144] A more successful approach is to use two equalizer stages, each one with a passive RC filter and a gain stage, introducing two zeros and two poles. The four corner frequencies can then be adjusted very delicately to optimize the flatness of the time-domain step response. It is this property that is needed to bring the very droopy receiver waveform of FIG. 35 up to the standard of FIG. 36.

[0145] FIG. 37 shows the circuit topology of a two-stage L-section filter, and the below equations express the form of the desired response of each stage:

A _≠ = A W _cql ^l and 4 _(/2 = A ₂ W _eq2 ^-^ .

S - S ₄ S - S ₆

The constants Ai, A ₂ , W _eq ι, and W _cq2 emerge from the complete derivation. The actual equivalent circuit is depicted in FIG. 38.

[0146] The total response of the system with the equalizers in place is given by

A - K A A W W ^S (S - ^S ₃ )( ^S - ^S ₅ )

(S - S ₁ )(S - S ₂ )(S - S ₄ )(S - S ₆ )

[0147] The barred frequencies Si and S ₂ are the internal poles of the transformer. The remaining corner frequencies are those introduced by the two-stage equalization network. A rather lengthy algorithm has been developed to optimize the time-domain step response in a least-squares fashion by adjusting parameters S ₃ - S ₆ . FIGS. 39 and 40 show the step response before and after optimization.

Advantages of the Gain Equalization Strategy

[0148] By using two stages of equalization with least-squares computation of the component values, the invention achieves a robust match of the EQ network to the transformer. Because there are redundant parameters in the nonlinear fit, it is possible to impart a quadratic sensitivity characteristic to the effect of component variations on the equalized pulse shape. Ultra-tight tolerances should therefore not be necessary to achieve consistent performance. Furthermore, the two cascaded gain stages provide a convenient way to achieve the amount of low-frequency boost required to minimize baseline wander.

[0149] It is important to note, that these embodiments are only examples of the many advantageous uses of the innovative teachings herein. In general, statements made in the specification of the present application do not necessarily limit any of the various claimed inventions. Moreover, some statements may apply to some inventive features but not to others. In general, unless otherwise indicated, singular elements may be in the plural and vice versa with no loss of generality. [0150] Although a specific embodiment of the invention has been disclosed, it will be understood by those having skill in the art that changes can be made to this specific embodiment without departing from the spirit and scope of the invention. The scope of the invention is not to be restricted, therefore, to the specific embodiment, and it is intended that the appended claims cover any and all such applications, modifications, and embodiments within the scope of the present invention.

[0151] What is claimed is: