An Experimental 10-Mbit/s Microwave Data Link

My own purpose in developing a 10-megabit radio link was threefold: to explore the design techniques underlying high-bandwidth microwave communications; to do my part to help bring Amateur Radio into the twenty-first century; and, most important of all, to use my company's T-1 Internet service at home for free. :-) This page is intended to describe my work to fellow Amateurs and other experimenters for whom a 9600-bps packet BBS or 31-baud PSK rig just doesn't cut it.

I've tried to describe the circuit and relevant construction techniques in depth, for those who may be interested in reproducing and/or modifying my work, but it should be made clear at the outset that such a project isn't a weekend endeavor for the electronics neophyte. While this isn't the most complex or demanding of projects, it has several unforgiving aspects that will not tolerate sloppy construction practices or inadequate care on the builder's part. You should be conversant with basic RF theory and construction skills before tackling it, or at least have access to one or two Elmers who have "been there and done that." You must have access to some basic electronic test equipment, including at a minimum a good oscilloscope and digital VOM. Finally, you should be prepared to commit the necessary resources of time and money to get the results you want. Underestimating the effort and expense required to get your project up and running is probably the #1 cause of failure.

System Architecture and Limitations

• Point-to-Point Gunnplexer Communications


The link as described uses two prepackaged microwave transceiver modules, called "Gunnplexers." These modules are remarkably simple and reliable. They consist of three diodes, a resonant cavity, and an antenna aperture. One of the diodes, the "Gunn diode," generates RF energy at 10 GHz. Another, the "varactor diode," is used to vary the exact frequency of the emitted microwaves in accordance with the Ethernet signal to be transmitted. Finally, the third diode, known as the "mixer," is responsible for receiving the signal from the Gunnplexer at the other end of the link. Because it's impractical to carry out the necessary Ethernet signal-recovery operations at the Gunnplexer's native frequency of 10 GHz, the mixer diode actually delivers the recovered signal at an "intermediate frequency," or IF, at a more reasonable frequency of 145 MHz. 

Where, exactly, does this 145 MHz intermediate frequency come from? It turns out that the mixer diode can mix, or heterodyne, the transmitted signal from the Gunn diode in its own Gunnplexer with the signal received from the distant end, to yield the desired IF frequency. All that is necessary is to tune the two Gunnplexers to frequencies which are 145 MHz apart. In our case, we arbitrarily designate one Gunnplexer for operation at 10.178 GHz, with its counterpart tuned 145 MHz higher at 10.323 GHz. The exact frequencies are not important (and are, in any event, almost impossible to measure precisely without microwave test equipment), as long as one Gunnplexer is tuned 145 MHz below its counterpart at the other end of the link. This heterodyne effect is essentially available for free as a consequence of the design of the Gunnplexer transceivers, and it saves us a great deal of trouble in the design and construction of the rest of the link's circuitry. 

But there are also a couple of key drawbacks to this simple idea. One is that there is no easy way to allow more than two Gunnplexers to communicate with each other. The use of three or more Gunnplexers in a multipoint network topology would be mathematically impossible, since there is no way that three or more signal sources can all be 145 MHz apart from one another! Consequently, the design as presented here is useful only for connecting two machines or subnetworks. 

• Operation without Ethernet Collison-Handling


Another consequence of the twin-Gunnplexer heterodyne technique mentioned above is that each of the two stations can receive data correctly only when it is not attempting to transmit data at the same time. Since the link uses frequency modulation (or more properly, frequency-shift keying) to transmit Ethernet data, a station which is transmitting packet data is not going to serve very well as a local oscillator for mixing with the transmitted signal from the other station. Our intermediate frequency of 145 MHz is really just a "center frequency" -- the ones and zeroes from the Ethernet port appear as rapid variations in the exact frequency over a range of several megahertz. At any moment in time, the IF frequency at a given station is the difference in the microwave frequencies being broadcast from that station and its remote counterpart. So if a station is transmitting data, it cannot simultaneously receive data from the other end because doing so requires that its own contribution to the IF frequency equation remains constant. This implies that Ethernet packet collisions will, in practice, be turned into lost or mangled packets. 

Fortunately, Ethernet turns out to be remarkably tolerant of such abuse. Collisions only happen in the first place when one station begins transmitting a packet after another station has inspected the state of the network and concluded that no traffic is currently being passed, in preparation for transmitting its own traffic. They are a normal part of life in the world of Ethernet, and generally become more and more frequent with increasing network traffic. When a collision is sensed by a fully-compliant Ethernet transceiver, the ultimate result is retransmission of the data packet that caused the collision in the first place. Retransmission also takes place in the TCP/IP protocol when a packet is corrupted or lost altogether, so the effect of our link's behavior is simply to turn one type of packet-loss condition into another. 

The distinction between lost packets and collisions is chiefly one of error-recovery performance. A lost packet is recognized as such only after a timeout period expires without acknowledgement of the packet's receipt at its destination. This period may range from milliseconds to seconds, according to the whims of TCP/IP and any other reliable-delivery protocols that the application may be employing. A packet which is involved in a collision can be caught instantly by an Ethernet-compliant transceiver and retransmitted in a more timely fashion, compared to one which is consigned to the proverbial bit bucket by our radio link. By failing to respond to packet collisions in a manner compliant with the Ethernet specifications, our link is theoretically vulnerable to performance losses that would not be a problem with a conventional hard-wired connection. In practice, however, with three machines at one end and eight at the other end of a single busy 10/100-megabit collision domain, we have not noticed any such degradation in either latency or throughput, despite running critical real-time applications on a near-constant basis. 

It's undoubtedly possible to design a separate signalling mechanism for reporting collisions to the AUI collision-sense input, along with the necessary circuitry to detect the occurrence of collisions in the first place. Perhaps the best solution for unusally collision-sensitive applications is to eliminate as many collisions as possible in the first place, by using a pair of switching hubs or bridges to partition the subnetworks at the two ends of the link into separate collision domains. To date, I have not run across any situations that warrant further action along these lines. 

• Connectivity with Ethernet Hardware via 10-Mbit/s Attachment Unit Interface (AUI)


It's tempting to press the ubiquitous 10BaseT interface provided on inexpensive network cards and hubs into service as the interface to a wireless data link. Unfortunately, though, another consequence of the Gunnplexer-based signal mixing technique is that transmitted data appears not only at the receiver at the remote end of the link, but at the local receiver as well! A few moments' thought will reveal that this is really the same problem that deprives us of a meaningful collision-sensing mechanism as described above. Because the intermediate frequency containing our recovered data is simply the difference frequency between the two transceivers, it doesn't matter which Gunnplexer does the transmitting -- the resulting frequency shift will appear in the IF signal at both stations. We're essentially caught in a no-man's-land between half-duplex and full-duplex communications. Although 10BaseT is physically capable of full-duplex operation, it's really a half-duplex signalling system, logically speaking. 10BaseT consists of two signalling pairs -- a receive pair and a transmit pair. Any activity on the receive pair while transmission is in progress is interpreted as a collision, even if that activity is just a locally-generated echo of the transmitted signal. Some Ethernet hardware may be configurable to overcome this limitation under particular circumstances, but in the general case 10BaseT is simply the wrong interface standard for a system such as this one. 

Enter the AUI port, or Attachment Unit Interface. Created especially as a general-purpose interface for Ethernet transceivers of every stripe, the AUI standard uses a DB-15 connector and comprises three signalling pairs: receive, transmit, and collision. The AUI port is ideal for our purposes, as it not only permits a locally-generated echo at its receive pair, but actually expects it. AUI signal levels and impedances are not greatly different from 10BaseT. AUI signals are somewhat lower in amplitude, and the Ethernet specification calls for shielded twisted-pair (STP) cabling to be used for connections to the AUI port, but I've had no problems connecting the transceivers to AUI ports with 15 to 20 feet (5 to 7 meters) of conventional Category 5 UTP cabling. The AUI ports's third (collision) pair is simply left unconnected. 

It's a little bit harder to find Ethernet cards and hubs with AUI ports, and they tend to be more expensive than similar gear equipped only with RJ-45 jacks for 10BaseT, but there are still plenty of bargains to be had. The prototype transceiver pair is connected to two LinkSys Enterprise Ethernet hubs featuring AUI ports and 16 RJ-45 jacks, purchased at CompUSA for about $150 each. In principle, 10BaseT could be used as the interface to the transceivers if a latching mechanism were designed to suppress the local echo, but the AUI port is definitely the way to go if possible. 

• Availability of an Unobstructed Line-of-Sight Path


This is a particularly important point to consider when planning your installation. Microwaves travel in a line-of-sight path and are easily blocked by buildings, large trees, or other landscape features. Additionally, large surfaces near or within the signal path may cause multiple delayed reflections of the transmitted signal to arrive at the receiver, in a manner similar to the "ghosting" sometimes observed on over-the-air TV signals. These multipath reflections, if strong enough, may cause data corruption. (On the other hand, a large planar surface such as the side of a building can also be used intentionally as a reflector if a good direct path proves impossible to achieve!) The transceiver units described here are equipped with large, easy-to-read signal strength meters to aid in optimizing the path, but keep in mind that multipath distortion can render even a strong signal unusable. 

Try to locate both transceiver units outdoors, so that the antennas face each other with no intervening obstructions such as windows. Many office windows are made with metallized coatings for energy-efficiency, which is bad news in the microwave business. One of my transceivers is located just inside a thick glass window that adds over 10 dB of loss to an otherwise-excellent 10 GHz path. In other words, 90% of the signal power is wasted on the trip through the window. If you find you must aim your transceivers through a window, try to keep the path as close to perpendicular to the window surface as possible to minimize attenuation. Transceivers located outdoors must be appropriately weatherproofed. This can be as simple as a layer or two of polyethylene sheeting or even a trash bag over the unit. Thin layers of plastic over the antenna aperture don't cause appreciable signal loss. 

For more information about signal-path planning and loss calculations, I'd recommend a look at The ARRL UHF/Microwave Experimenter's Manual. Select the "VHF/UHF/Microwave Communications" link on this page for a description of this and other good references. 
 
 

Design and Construction Notes

• Antenna Considerations


No radio is better than its antenna. Adding a better antenna to a communications system is analogous to putting a race car on a weight-reducing diet -- it's the single best way to increase performance in almost every parameter. Before you increase the power of your transmitter, use exotic devices to optimize your receiver's noise figure, or otherwise throw money or engineering effort at a signal-strength problem, you should first make sure your antenna system is the best it can be. Fortunately, one of the biggest advantages offered by microwave radio communications is also one of the smallest ones -- antenna size! It's not difficult to achieve effective radiated power (ERP) levels of several watts or more using 10-milliwatt Gunn diodes, since a 1- or 2-foot dish antenna is electrically "large" at these ultra-short wavelengths. You can reasonably expect reliable operation of the radio link over several miles if proper attention is paid to antenna design and aiming.

Although dish antennas are clearly the best choice when maximum range is the goal, they aren't likely to be necessary for links which span less than a mile (1.6 km) or so. The prototype transceivers use homemade horn antennas fabricated according to a 1:1-scale template generated by Paul Wade, N1BWT's excellent HDL_ANT antenna-design program. Here are two ready-to-build designs suitable for use with the M/A-Com 10 GHz and 24 GHz Gunn transceivers, respectively. Initial testing and alignment of the link can be accomplished with no antennas at all, but unless you are contemplating extremely close-range work (less than a hundred feet or so with no obstructions) you will want to fabricate horn antennas for your transceivers. You should observe about 5 dBi (i.e., 5 dB of gain over an isotropic radiator in free space) with the "naked" Gunnplexer apertures alone, and about 8-10 dB of additional gain with the 15 dBi 10 GHz horn mentioned above. To construct each horn as shown in the photo, I used sheet brass obtained from a local hobby shop, scoring the inner seams with a Dremel tool to allow the antenna to be folded together origami-style, and soldered the remaining seam. There were plenty of leftover scraps available for mounting the antenna to the Gunn transceiver aperture. It's important to achieve a tight fit between the antenna and aperture -- don't let the small end of the horn "float" too far off the Gunnplexer's mounting surface. N1BWT's latest page has a great deal of knowledge in store for those who are interested in optimizing microwave antenna performance, especially his online Microwave Antenna Book.

The financially well-endowed may choose to go with a commercial standard-gain horn antenna such as the HWR-90 model sold by Microtech, Inc., but their price of U.S. $400 each will likely send most experimenters reaching for their dust mask, safety goggles, and Dremel tool instead. 

• Microwave Transceiver Modules


The prototype transceivers were fabricated with two 10 GHz M/A-Com MA87728-M01 voltage-controlled oscillator transceiver modules. I'm referring to these with the older term "Gunnplexer," for simplicity's sake. These modules were obtained from SHF Microwave Parts Company at a cost of about US $90 each. These modules deliver excellent performance, are easy to work with, and can be upgraded from 10 mW output to 100 mW by installing optional high-power Gunn diodes available at US $45 each. Although there has been some uncertainty regarding the continued availability of the 10 GHz Gunnplexers, the latest news on the SHF Microwave Parts site (as of August 13, 1999) indicates that their distribution issues with M/A-Com have been resolved to everyone's satisfaction. According to Alan Rutz, WA9GKA at SHF Microwave:

1800 hrs, Chicago time, 13 Aug 99..well, we are back in business with M/A-Com. The decision was made just a few minutes ago, and we will once again be receiving both the 10 ghz and the 24 ghz M/A-Com Gunn sources... What's better, we will have our next shipment already next Tuesday, so we will be able to ship right away. Prices may change up or down by just a little, and our price will always be shown in the advertisement below. Expect small price changes soon. Thanks for your kind support! 

In addition to the 10 GHz devices mentioned above, SHF Microwave also distributes 24 GHz Gunn transceivers. These units are essentially identical to the 10 GHz devices used in the link, and can be substituted with no circuit modifications. They are less expensive as well. However, they are quite a bit smaller and somewhat less powerful (5 mW versus 10 mW) than their 10 GHz counterparts. I used a pair of the 24 GHz Gunnplexers in my initial experiments while developing the radio link earlier this year, but found that the signal they emitted was attenuated too severely by the glass window at the "office" end of the link. The 10 GHz Gunnplexers had no problem putting a good signal through the glass, especially after I upgraded them to the 100 mW Gunn diodes! Not being especially talented at metalworking, I also found it easier to fabricate the larger horn antennas for the 10 GHz units. Finally, there are plenty of reasonably-inexpensive spectrum analyzers available on the surplus market, including my own Tektronix 492, that make it trivial to set the 10 GHz Gunnplexers to any desired frequency. 24 GHz spectrum analyzers, on the other hand, still seem to cost about as much as a new Porsche. All in all, I prefer the 10 GHz Gunnplexers to the 24 GHz parts and recommend them as a first choice if you don't have any particular reason for wanting to run your link on 24 GHz. From what Alan is reporting, it appears that both models will continue to be available in the foreseeable future.

Data sheets for the 10 GHz and 24 GHz Gunnplexers are available here. Alan has annotated the 24 GHz model's data sheet to clarify the misleading pin nomenclature used by M/A-Com. As with any sensitive microwave semiconductor device, it's vital to keep static electricity away from the various Gunnplexer diodes, especially the mixer diode. These units come with shorting straps that should not be removed until you've soldered the 2.2K mixer load resistor into place between the mixer diode and ground terminals. Don't apply power until you've checked and double-checked your pin connections!

I will attempt to keep this page up to date with any new information that comes to light on the availability of suitable transceiver modules. Please email me if you're aware of any other sources for Gunn transceivers. (The cheaper surplus models that don't include varactor diodes are not suitable for use at 10 megabits per second, as they lack the necessary high-speed FM modulation capabilities.) 

• IF Strip


Our intermediate-frequency (IF) amplifier circuit is not unlike the one used in Glenn Elmore, N6GN's original 2-MBit/s design. Two Mini-Circuits MAR-6 monolithic microwave integrated circuits (MMICs) are cascaded to yield approximately 40 dB of gain at 145 MHz. Where the N6GN design used a set of low-pass and high-pass filters constructed with discrete capacitors and inductors to provide the desired IF bandpass response, I've used the factory-made PHP-100 and PLP-200 filters from Mini-Circuits for a wider IF response and easier assembly. Together, these two filters result in -3 dB IF response skirts at 82 and 210 MHz -- more than wide enough to pass the modulated Ethernet spectrum without introducing excessive phase distortion.

While the Mini-Circuits filters are relatively inexpensive, they could easily be replaced with homemade filters similar to N6GN's, if you have the equipment to test them. Several basic filter-design programs can be found on the net, including Bob Lombardi's FilDes for MS-DOS.

One important distinction between the IF strip shown here and the N6GN design is the presence of a Mini-Circuits T4-1 broadband transformer between the Gunnplexer's mixer diode and the first MAR-6 amplifier. The purpose of this transformer is to provide a better match between the relatively high-impedance mixer diode and the 50-ohm MMIC input. Without it, the effective receiver noise figure is degraded by several dB due to excessive mixer-diode loading. The difference in weak-signal recovery performance is dramatic. It's surprising that the original N6GN article didn't include this useful tweak. 

• FSK Demodulator


Based on the MC13155 wideband FM IF subsystem IC from Motorola, the FSK demodulator circuit follows the IF strip. Its purpose is to recover the original Ethernet waveform from the 145 MHz IF signal. A Mini-Circuits MAR-4 MMIC and T4-1 broadband balun deliver the IF signal to the MC13155, where its exact frequency at any given instant is converted into a balanced differential output representing the amplitude of the Ethernet signal being used to modulate the Gunn varactor diode at the opposite end of the link.

Besides the MC13155, the other key component in the FSK demodulator is the quadrature tank circuit. This is a parallel LC resonant circuit (visible to the right of the MC13155 IC in the picture) which is tuned to the 145 MHz IF center frequency. As the FSK-modulated IF signal varies above and below the center frequency, the quadrature tank circuit exhibits a varying amount of phase shift which the MC13155 interprets to determine the correct output logic level. Because the 10-Mbit/s data rate requires a high signal bandwidth for efficient transmission, it's necessary to deliberately lower the Q factor of the quadrature tank circuit to approximately 3.6 by "swamping" it with a 180-ohm resistor. The resulting demodulation bandwidth is given by (145 / 3.6), or approximately 40 MHz. A signal near 125 MHz will yield a low output level from the MC13155, while a signal near 165 MHz will result in a high output level. The center IF frequency of 145 MHz corresponds to the baseline Ethernet signal level which is present when no data is being transmitted.

Guidelines for alignment and testing of the FSK demodulator circuit appear below. 

• Output Buffer


Once the Ethernet signal has been recovered by the MC13155, it must be level-shifted, filtered, and amplified prior to delivery to the AUI port. The MC13155 is operated in a positive-ECL (PECL) supply configuration, resulting in output levels in the vicinity of 3.4 volts for a "low" signal and 4.2 volts for a "high" signal. In the output buffer circuit, a Burr-Brown OPA603 wideband current-feedback op-amp acts as a level shifter, bringing the PECL signal to a baseline value near 0V DC with a low-impedance output suitable for driving a Mini-Circuits PLP-21.4 low-pass filter. This filter's job is to remove any residual traces of the 145 MHz IF signal from the FSK demodulator. Without it, excessive IF signal leakage superimposed on the recovered Ethernet waveform may cause data corruption and degrade the link's signal-to-noise performance.

The filtered signal is then applied to a second OPA603 for amplification to a level which is compatible with the Ethernet AUI port. Note that this is still an "analog" signal in every sense of the word -- it has not passed through any comparators or logic gates on either the transmitter or the receiver side of the link. Keeping the Ethernet signal in the analog realm by processing it with linear circuitry ensures that we are able to preserve all of the edge characteristics of the signal as it transitions to and from the "neutral" state at the beginning and end of each packet. Experience has shown that Ethernet devices can be finicky about edge-preservation, so our intention is to avoid potential edge-handling complications by leaving the job of digital signal conditioning up to the Ethernet hardware itself.

The output buffer circuit is fairly straightforward, with the exception of the 938-ohm resistor that controls the DC offset at the output of the first OPA603 op-amp. As the schematic's notes suggest, this resistance value can be obtained with a parallel combination of 1K and 15K resistors, or with a 10-turn precision trimmer. In either case, a good digital VOM can help you obtain a resistance value close to 938 ohms, either by hand-selecting 1K and 15K resistors which yield a total resistance within 1% of the desired value or by presetting the trimmer adjustment. The idea is to position the output Ethernet signal at pin 6 of the final OPA603 to a baseline value close to 0V DC when no signal modulation is present. Keeping the average DC voltage near 0V helps avoid core saturation in the Mini-Circuits T1-1 broadband transformer which is used for isolation between the transceiver and Ethernet device. Small changes in the 938-ohm resistor's value can yield large shifts in the average DC output level, so it makes sense to strive for precision in this part of the circuit. You may wish to verify the DC offset of the output waveform with an oscilloscope, to make sure the baseline level is within 100 mV or so of 0 VDC. The same unmodulated 145-MHz signal source used to align the FSK demodulator (below) can make the output buffer offset easy to tweak as well.

In any event, you should resist the temptation to use AC coupling in lieu of the DC level-shifting components in either the output buffer or FSK modulator circuits. Ethernet packet traffic appears as a series of periodic signal bursts with widely-varying duty cycles. Capacitors used for AC coupling between stages with different DC operating points (such as between the MC13155 and the output buffer amplifier, or between the AUI transmit circuit and the Gunnplexer varactor diode) inevitably introduce an RC time constant characteristic which can cause the average DC signal level to vary over time, both between and during packet bursts. Because of the aforementioned sensitivity of Ethernet hardware to the integrity of signal edges at the beginning and end of each packet, any component which introduces a time-variant DC characteristic is bound to be a bad thing.

In fact, this effect cost me several weeks of head-scratching and an astonishing amount of money during development of the prototype link. I found I could receive either low-duty-cycle pings or continuous high-intensity traffic near 10 Mbits/s, depending on the AC-coupled circuit characteristics -- but I was never able to find a configuration that would allow both types of traffic to pass! Given the complexities of choosing the filters, amplifiers, signal levels, deviation and bandwidth figures, and other operating conditions at both ends of the link, this was one problem I did not need. (Most of the R&D budget for the project went straight to Tektronix for one of their nifty new digital-phosphor oscilloscopes, which, in desperation, I convinced myself I needed in order to track down the link's signal-integrity problems. :-)

It should also be noted here that the AUI ports on the Ethernet hubs I used were not enabled by default. Some experimentation revealed that the hubs would not recognize an attached AUI transceiver unless all of the ground pins on the AUI port were tied together to signal the presence of a device plugged into the port. As the schematic indicates, pins 1, 8, 11, and 14 should be bonded together at the AUI plug. Because the AUI signals themselves are differential in nature, this ground connection does not need to be connected to the transceiver ground at either end. I have heard vague rumors that some manufacturers' AUI ports also expect to see at least a few milliamps' worth of loading at their +12V output pins in order to recognize the presence of a connected transceiver device. This was not the case with the LinkSys hubs I used, but may be worth investigating further if you are unable to get the link working in your case. It is easy enough to verify AUI-to-AUI connectivity with a 4-wire crossover cable (pins 3-5 and 10-12 joined at both ends) if you have doubts about your ability to communicate with your particular hubs or NICs. Please email me if you are able to verify or dispute any of these observations with respect to the AUI interface. 

• FSK Modulator


At first glance, the FSK modulator circuit is not unlike the output buffer in operation, from a mirror-image perspective. Its purpose is to amplify the signal representing the Ethernet data to be transmitted, while shifting its average DC level from a zero-volt baseline to a desired positive voltage without introducing any time-variant envelope distortion. (In short, all of the cautionary notes regarding capacitive coupling in the output buffer apply to the modulator circuit as well!) The amplified and level-shifted Ethernet signal is applied to the Gunnplexer varactor diode, where it imparts FM modulation to the microwave transmission by causing rapid fluctuations in the electrical characteristics of the Gunn diode's resonant cavity.

The Ethernet signal to be transmitted enters the modulator circuit through a Mini-Circuits T2-1T broadband transformer, configured as a balun to provide isolation and an improved impedance match between the 78-ohm AUI port and the Mini-Circuits PLP-21.4 low-pass filter which follows it. By Carson's Rule, the bandwidth occupied by a transmitted FM signal is related to the highest modulating frequency present. Consequently, the low-pass filter is employed to eliminate high-frequency energy present in the Ethernet waveform which would otherwise waste bandwidth without carrying useful information.

After filtering, the Ethernet signal is amplified by a Burr-Brown OPA603 wideband current-feedback op-amp. The variable resistor R1 controls the amplifier's gain, which determines the amplitude of the output waveform applied to the Gunn varactor diode and, consequently, the deviation of the transmitted FSK signal, or the magnitude of its excursions from its center frequency. At the OPA603's output, a 1N4148 diode provides cheap insurance against possible damage to the Gunnplexer varactor caused by forward-bias transients when powering the link up or down.

When designing a frequency-shift-keyed transmitter such as this one, the desired FM deviation is typically chosen on the basis of the highest modulating frequency expected to be transmitted. The deviation ratio, given by the ratio of these two values, is not particularly critical, as long as it is sufficient to generate intelligence-bearing sidebands of reasonable amplitude, without occupying an unnecessarily-large chunk of bandwidth. A deviation ratio between 1 and 2 is typical for such applications. One reason the exact deviation ratio isn't critical is that a precise, optimal transmission bandwidth budget is tricky to calculate for a complex waveform such as an Ethernet signal, or even a voice transmission. Prior to being used to modulate an FM transmitter, most of the energy in a 10-Mbit/s Manchester-encoded baseband Ethernet signal exists around the 10 MHz point, falling off several megahertz to either side. Very little signal-bearing energy remains by the time our 21.4-MHz low-pass filter cutoff frequency is reached, so the filter's purpose is largely to eliminate harmonics generated in the process of forming the Ethernet waveform.

For those interested in a more detailed look at the subject, W6QS has posted an excellent, easy-to-follow graphical model of 10-Mbit/s Ethernet and its FSK modulation characteristics at the Tech Bench Elmers Amateur Radio Society page.

Finally, an actual spectrum-analyzer screen shot of the FM-modulated 10 GHz signal from one of the prototype transceivers appears here. This particular snapshot was taken while several megabytes' worth of MP3 files were being copied from one end of the link to the other at full Ethernet speed.

• Tuning and AFC Considerations


Because the frequency of the signal transmitted by each Gunnplexer is determined by the voltage level applied to its varactor diode, the average DC level at the output of the FSK modulator circuit can be thought of as the "tuning control" for the link. The DC bias applied to the varactor diodes at both transceivers ultimately determines the 145 MHz IF separation. Consequently, a mechanism is needed to ensure that the nominal transceiver frequencies stay locked 145 MHz apart from each other regardless of temperature variations and component aging characteristics that could cause the frequencies of the individual Gunnplexers to drift out of alignment.

Fortunately, because the relative frequency separation between the two transceiver units is more important than their absolute frequencies, any such drift-compensation circuit needs to be implemented at only one end of the link, not both. In our case, the output buffer circuit at the "low-frequency" (10.178 GHz) end of the link has a few extra parts. A low-gain amplifier consisting of one stage from an LM324 quad op-amp is connected to monitor the differential outputs from pins 4 and 5 of the MC13155 FSK demodulator, with its output fed via a low-pass RC integrating filter and bipolar transistor buffer to a voltage-summing junction in the FSK modulator circuit. The OPA603 in the FSK modulator circuit adds the constant voltage at its inverting input to the output signal it delivers to the Gunnplexer varactor. (Because the sign of the added voltage is inverted by the OPA603, we apply a negative voltage to the summing junction to yield a positive bias offset at the varactor.)

The result is a simple feedback loop, in which any difference in the DC level between the two FSK demodulator output pins is amplified, integrated over time, and used to steer the IF difference frequency in the proper direction to cancel out the unwanted offset voltage. The filter circuit, consisting of a 4.7 uF capacitor and a 10K series resistor, is necessary to keep the AFC loop from trying to cancel out the desired rapid changes in frequency arising from the normal Ethernet modulation process.

At the other end of the link -- the "high-frequency" end which operates at 10.323 GHz -- we simply feed the modulator summing junction with -8 volts through a fixed 820-ohm resistor, resulting in a constant DC offset of about +5V. This is enough to bias the varactor diode into a reasonably-linear portion of its voltage-versus-frequency characteristic curve, while providing sufficient headroom on both sides for the applied Ethernet modulation. If the Gunn varactor diode or its associated driving circuitry exhibits any drift, the AFC circuit at the other end of the link will adjust its own Gunnplexer's center frequency to compensate. The distinction between the two transceivers is arbitrary -- note that we could just as easily have applied AFC correction at the high-frequency transceiver unit instead, swapping the connections between the MC13155 and LM324 to change the sign of the correction signal.

With one transceiver mounted indoors and kept at a relatively constant temperature of 75 degrees F, this AFC mechanism has proven capable of maintaining the integrity of the link at Gunnplexer temperatures between approximately 50 and 140 degrees F as measured at the outdoor end of the link. Beyond 140 F, the system ran out of compensation range and failed. The excessive heat buildup was due to the effect of the outdoor unit's near-airtight RF shielding on a particularly warm day, aggravated by the use of plastic wrap for rainproofing. A small fan was installed in the transceiver's outer housing to aid ventilation, eliminating further problems. 

• Tuning/Fault Meter Circuit


The AFC circuit in the low-frequency transceiver unit includes an additional LM324 buffer section which is used to amplify the differential voltage from the MC13155 demodulator outputs to drive an analog panel meter. In operation, this meter should always read near center scale, indicating that the average IF frequency is on target at 145 MHz. A problem with the AFC system or related circuitry that causes the two Gunnplexers to drift out of alignment with each other will result in a constant tuning-meter deflection to one side or the other.

In the prototype unit, this meter actually serves as a power-supply current monitor in addition to monitoring the AFC circuit. The meter, installed at the link control box, uses the 15' power-supply ground line as its DC signal path return. Instead of a conventional zero-center tuning meter as a frequency indicator, I used a 0-100 uA full-scale meter, with a separate sensitivity control that adjusts the meter for a center-scale reading under normal power-supply load conditions. Any fault condition with the Gunn diode or other portion of the system with substantial current requirements will deflect the meter off-center. This dual-purpose monitoring scheme was a good match for the full-scale 0-100 uA meter I had on hand, since the current drain from the optional 100 mW Gunn diode I used was sufficient to drive the meter to center scale under normal operating conditons. In reproducing the project, you may elect to use a standard zero-center tuning meter with a separate signal ground line, or even omit the meter circuit entirely. 

• S-Meter Circuit and Signal Strength Interpretation


Both the low-frequency and high-frequency transceiver units include a received signal-strength indicator (RSSI) circuit which displays the strength of the received signal on a digital panel meter at the link control box. The MC13155 FSK demodulator circuit provides a logarithmically-detected RSSI output at pin 12 which ranges from 0 to approximately 75 millivolts depending on the amplitude of the 145 MHz signal from the IF strip. This low-voltage output is amplified by an LM324 buffer and sent to the link control unit at a higher voltage for improved noise immunity.

The prototype transceivers use a pair of Extech 390GVS2 meters, available from Future-Active Electronics. This digital panel meter is designed for direct readout of applied voltages between 0 and 200 mV, so a resistive voltage divider is used to bring the amplified voltage from the S-meter buffer back down to the meter's display range. The resistor values shown will yield a displayed voltage within a few millivolts of the actual RSSI output from the FSK demodulator subassembly.

In one prototype unit, the following S-meter values were observed and charted next to the actual voltage measured at the FSK demodulator subassembly's RSSI output connection. The negative S-meter readings at low RF levels are due to a combination of the LM324 buffer's offset voltage and the voltage drop along the transceiver power cable's ground lead, the latter being dominated by the substantial negative current consumed by the Gunn diode. The input signal was an unmodulated 145.0 MHz carrier applied at the input of the MAR-4 IF buffer amplifier.

RSSI levels and S-meter readings versus CW input power

		Signal Level
		Displayed S-meter indication	Actual RSSI voltage in mV
IF power in dBm	(No signal)	-5.0	2.4
	-60	-4.6	2.8
	-55	-3.6	3.8
	-50	0.0	6.8
	-45	8.0	15
	-40	20	25
	-35	28	33
	-30	38	42
	-25	50	53
	-20	60	63
	-15	65	68
	-10	68	70
	-5	70	72
	0	72	74

The prototype's threshold of data recovery was approximately -45 dBm at the FSK demodulator subassembly input. In general, any S-meter indication below 20 or so corresponds to a signal level which is too low for reliable operation. The signal path depicted in the photo yields S-meter readings around 35-40 with the optional 100 mW Gunn diodes installed in the M/A-Com 10 GHz transceiver modules, suggesting a usable S/N margin of about 10 dB. (The IF strip's gain of 40 dB can be added to the levels shown above to determine the signal available at the Gunnplexer mixer diode for a given S-meter reading.)

Like the tuning/fault meter circuit, the exact design of the signal-strength monitoring circuit can be left to the builder's discretion. Omitting the S-meter circuit isn't recommended, as it is almost indispensible for antenna aiming and link diagnostics. 

• Power Supply, Shielding, and Decoupling Considerations


Each transceiver unit will need to be equipped with a power source capable of supplying +12VDC and -12VDC at around 1A. This can take the form of a simple homebrew linear supply design, or a ready-to-install commercial power supply module. The prototypes use a pair of TDK MRW170KV switching power supplies obtained from Vetco Electronics, a Bellevue, Washington surplus dealer. The TDK supplies can deliver up to 2.5A at +/- 12V and 7.0A at +5V -- more than enough current to power the link without breaking a sweat. They appear to be identical to the Kepco models which bear the same part number, and at only $19.95 each from Vetco, it's hard to beat the price.

Apart from the output voltages available from the main power supply, the transceiver subassemblies require power busses at +9, +/-5, and -8 VDC. These voltages are easy to obtain from 3-terminal 78/79xx regulators driven from the main power supply. (The +5V supply to the MC13155 is derived from a dedicated 7805 regulator inside the FSK demodulator subassembly). To keep the diagrams simple, none of the 3-terminal regulators and their associated bypass capacitors are shown on the schematics, although they can be seen in the overall view of the prototype transceiver assembly if you look closely. Two regulators are mounted on the cover of the IF strip subassembly, with the remaining pair mounted to either side of the 9-pin D-sub connector where the power supply and signal-monitoring cable from the link control unit plugs in.

Because the microwave link is a broadband system with very high gain, good RF shielding and decoupling practices are quite important when assembling the transceiver units. While it's not necessary to follow the prototype examples faithfully in every respect, a few general suggestions are worth pointing out.

• Use subassembly construction techniques.

This practice not only helps keep internal RF signals where they belong, but also makes it easy to assemble and test individual modules independently of their neighbors. Not only that, you'll have the option of reusing entire modules in future projects, instead of salvaging individual parts! Although it does add to the effort and expense of construction, I am a die-hard believer in enclosing all but the most trivial RF circuits in Hammond 1590-series boxes. These enclosures are available in a variety of sizes from almost every mail-order vendor including Digi-Key and Mouser. The IF strip and FSK demodulator subassemblies were built in Hammond 1590A boxes which were themselves enclosed in a larger 1590D chassis. The FSK modulator circuit and RJ-45 jack used for the AUI cable connection were installed in a separate 1590B enclosure which was bolted to the exterior of the main chassis box.
• Bypass all power-supply and monitoring lines at the point where they enter the main enclosure.

Each transceiver's power and tuning/S-meter lines are brought into the outer enclosure via a 9-pin D-sub male jack. Pairs of good RF-quality 0.1 uF ceramic capacitors and 4.7 uF tantalum capacitors are used to bypass the +/- 12V lines at the connector itself, while the S-meter and tuning-meter signal lines are bypassed with 0.1 uF capacitors alone. The reason for bypassing at the point of entry, as opposed to any other point within the enclosure, is that the length of wire between the D-sub connector and the bypass point will act as an antenna, radiating stray RF signals from paging, FM, and VHF television stations into the supposedly RF-tight enclosure! Even a few centimeters of unbypassed connecting wire can couple enough energy into the enclosure to degrade the link's sensitivity by several dB. If either of your transceivers shows more than 5-10 millivolts at the S meter with all enclosure covers in place and the other unit powered off, you should suspect inadequate shielding or bypassing. An excellent tutorial on RF shielding practices can be found at Wes Hayward, W7ZOI's site.
• Treat high-bandwidth, high-gain components with respect.

Components such as the Mini-Circuits MMIC amplifiers, OPA603 opamps, and MC13155 FSK demodulator are nothing short of miracles of solid-state integration. A few short years ago, parts like these simply could not be manufactured, much less purchased for the price of a nutritious lunch for two at McDonald's. Although these three examples are devices with widely-differing purposes and specifications, they all share a common trait which may not be obvious at first glance: very high gain over a bandwidth extending from DC into the hundreds of megahertz. Don't even think about mounting these parts on conventional "perfboard" or Vector-style boards that lack an extensive ground plane. Keep lead lengths short, and pay attention to places (e.g., the OPA603 input pins) where it's actually best to leave some clearance to the ground plane in order to minimize stray capacitance. As the photographs will reveal, my personal sympathies lie with the "dead-bug" or "ugly"-construction school of thought, in which components are suspended by their own leads over a ground plane at the surface of an unetched copper-clad board. Conventional PC boards are fine as well, as long as their layouts and ground planes are carefully designed.

Perhaps the most important advice of all is to study the manufacturer's data sheets and app notes for the parts you plan to use, even if you don't understand everything being discussed. The manufacturer's application engineers have seen every problem under the sun, and it's in their interest to warn you about potential complications that are likely to arise when working with their chips.

Alignment and Testing

• Gunnplexer Frequency Calibration


Begin by arranging to supply the required power to both units' Gunn diodes, with at least 0.1 uF of bypassing at the Gunn diodes' terminals. Pay close attention to the supply voltage, current, and polarity requirements of the Gunn diodes in the units you are using. Don't operate a Gunn diode at less than its rated voltage -- they are negative-resistance devices. Continued operation at a lower voltage may actually increase the device's temperature to an unsafe level.

In soldering to the Gunnplexer diode terminals, it's best to use a very hot iron with a small, well-tinned tip to keep from "soaking" the diodes with excessive heat. A low-wattage iron brings with it a greater risk of damage to solid-state devices, since it will take longer to make a good connection. A grounded iron suitable for use with static-sensitive devices is ideal, if one is available.

Next, apply +5V DC to each Gunnplexer's varactor diode to establish the voltage corresponding to the center IF frequency. It's best to feed this voltage through a temporary network consisting of a 10K resistor followed by a 0.1 uF bypass capacitor, to keep power-supply noise from complicating the alignment process. If you have access to a microwave spectrum analyzer or frequency counter, simply adjust the Gunnplexers' mechanical tuning screws to set one unit to 10.323 GHz and the other to 10.178 GHz. Again, the exact frequencies are not critical as long as they are within a few MHz of 145 MHz apart. Keep track of which unit is which -- you'll need to install the 10.178 GHz Gunnplexer into the "low-frequency" transceiver to maintain the correct Ethernet signal polarity.

If you don't have a way to measure the Gunnplexer output frequencies directly, you'll need to monitor the mixer output of one unit to measure the difference in their frequencies. With both mixer-diode shorting straps in place, carefully solder a 2.2K resistor between the mixer terminal of one Gunnplexer and its ground terminal. The shorting strap on this unit may now be removed. Separate the two Gunnplexers by a few inches, with their apertures facing each other, and connect a VHF frequency counter, spectrum analyzer, or receiver to the mixer terminal. Slowly tighten either Gunnplexer's tuning screw into its resonant cavity until a difference frequency of 145 MHz is received. This will appear as a quieting signal on a 2M receiver. To make sure you're in the 10.0 - 10.5 GHz amateur band, you may wish to tighten the screw on the other Gunnplexer at this point until the same result is achieved. Clockwise rotation of the tuning screw will lower the transmitted frequency. Since the Gunnplexers' initial frequency as delivered from the factory is close to 10.525 GHz, this two-step process should put your units safely within the ham band. Should you find your units hopelessly misaligned, an inexpensive police radar detector with X-band capability can help you retune them to a frequency near 10.525 GHz. Radar detectors typically respond to a range of frequencies around 60 MHz either side of 10.525 GHz, so you should be able to achieve a good center-frequency response with the tuning screws.

As an alternative means of alignment, if you have access to a 145 MHz signal source (either a 2M rig or a calibrated RF signal generator) you may find it more convenient to build and align the FSK demodulator assemblies first. Temporarily connect one of the demodulator assemblies to the mixer diode test point. (The IF strip will not be needed for close-range testing.) The output voltage between pins 4 and 5 of the MC13155 will pass through 0V as the difference frequency between the two Gunnplexers passes through 145 MHz. In fact, this makes a good sanity check even if you have already aligned the Gunnplexers by other means. Watch out for false nulls -- the response at the correct adjustment point should be sharp and symmetrical to either side.

• FSK Demodulator Calibration


The LC quadrature tank circuits in the two FSK demodulator subassemblies must each be hand-adjusted to a resonant frequency of 145 MHz. As with the Gunnplexer output frequency adjustments, absolute frequency accuracy is not as important as achieving a near-perfect match between the two transceivers. A good starting point is to fashion your tank-circuit inductors to resemble the one in the prototype's photo as closely as possible. Wind 3 turns of #22 enamelled copper wire (such as Radio Shack catalog #278-1345) around a 1/4" (6 mm) form such as a screwdriver shaft, then withdraw the form and spread the turns out so that the coil's aspect ratio resembles the example. (A slug-tuned inductor can also be used, provided the core material is suitable for use at VHF frequencies.) Install the coil and its associated 22-pF capacitor and 180-ohm resistor as close as possible to the MC13155 IC. Ideally, the coil should be separated from large, permeable surfaces such as the subassembly enclosure's wall by at least its own diameter.

An accurate 145 MHz signal source should be used for final alignment of each FSK demodulator circuit. This can be a 2M transmitter operating on the bench a few inches away from the circuit under test, a calibrated signal generator, or even an aligned pair of Gunnplexers. The latter option can be especially attractive since a good frequency match between the Gunnplexers' IF difference frequency and each FSK discriminator's tank circuit is exactly what we're trying to achieve.

With the signal source turned on, use a digital VOM or oscilloscope to monitor the differential voltage between pins 4 and 5 of the MC13155. It's easy to connect a monitoring instrument if you route these signals to a pair of RCA jacks, as in the prototype. Spread or compress the coil's turns by hand, using a toothpick or other nonmetallic alignment tool, in order to minimize the voltage difference between the two pins. Your goal should be to come within +/- 10 to 20 millivolts of 0V. Fine adjustment can be accomplished by physically bending the leads of the 180-ohm resistor or 22-pF capacitor to move the component's body back and forth relative to the coil.

Finally, remove the alignment tool and install the FSK demodulator subassembly's cover to ensure that the coil's tuning characteristics don't change drastically with the cover in place. If a marked change in alignment occurs, remove the cover and separate the coil's turns slightly to compensate.

If you have a 50-ohm signal generator with a calibrated output attenuator, you may wish to check the amplitude response at each demodulator assembly's RSSI output against the table above. RSSI output values within +/- 20% should be noted at each input level setting.

• FSK Modulator Deviation Adjustment


The modulator deviation adjustment (R1 in the modulator circuit) is not especially critical. I've chosen to use a 500-ohm carbon (not wirewound!) potentiometer in the prototypes, but it's not unreasonable to install a 220-ohm fixed resistor in its place if you're using the recommended 10 GHz M/A-Com Gunnplexers. With R1 set to 220 ohms, the OPA603 amplifier's voltage gain is roughly 3.5X, as demonstrated by its input and output waveforms. These oscilloscope screen shots were taken at the AUI port and Gunnplexer varactor diodes, respectively, and depict not only the gain of the OPA603 modulator stage but the insertion loss and high-frequency rolloff effects of the PLP-21.4 low-pass filter as well.

Links which are constructed with substantially different microwave transceiver modules, or which need to be carefully optimized for maximum S/N performance, should employ a variable resistor at R1 to simplify calibration. If you have an oscilloscope with a 50-100 MHz bandwidth, it would definitely be a good idea to verify that your LO drive waveform is similar to the prototype's in amplitude and DC offset with actual Ethernet data being transmitted. Since I do not have a copy of the official Ethernet specification to work with, I'm assuming that wide variations in AUI output amplitude are not likely to show up among different manufacturers' hubs and network interface cards! As always, I'd appreciate any feedback regarding this issue.

• Testing and Troubleshooting


Once the adjustments above are made, the link should be up and running! Here are a few hints for sorting out any problems that may arise. I'll be glad to try to help if you email me with questions or problems, but please be prepared to provide details with regard to what you've already tried and what observations you've been able to make. "It doesn't work!" is not a valid appeal for help. :-)

• Test each subassembly by itself, when possible. You can approach a "dead link" problem with confidence if you are able to validate the operation of each part of the circuit independently of its surrounding modules. Ideally you'll know your Gunnplexers are on frequency, because you're looking at their output signal with a microwave counter or spectrum analyzer. You know the IF strip is delivering close to 40 dB of gain in a stable manner, because you're able to feed it with a 150 MHz signal generator of known amplitude and observe its parasitic-free output with a VHF/UHF-capable spectrum analyzer. Your FSK demodulator subassembly's RSSI output levels are comparable to those shown in the table at varying input levels from a calibrated signal generator, and the AUI input and output buffer circuits yield waveforms similar to the prototype's with either an injected 10 MHz signal or actual Ethernet data. If you lack access to the basic test equipment needed to verify operation at the module level, your troubleshooting job may become harder than it needs to be. It can pay to make friends with some grad students or faculty members at your local university's EE department, or the staff at a local lab that provides FCC certification services. Local ham radio clubs may have members who are able to help with testing as well.
• Test each transceiver unit by itself in "loopback" mode. The local-echo characteristic of the Gunnplexer-basd transceiver architecture can be an extremely valuable troubleshooting aid. It's much easier to get the link stations up and running one at a time! Simply build an AUI crossover cable to connect your hubs directly, and splice one of the transceivers into one of the wire pairs. The other transceiver's AUI connection port is left unconnected. If your connection phasing and transceiver alignment are correct, the Ethernet signal present at the input of the transceiver under test should be echoed to the output of the same transceiver. Once both transceivers pass this loopback test, they should be able to "talk" to each other without difficulty. This would be a good test to perform after verifying that your transceiver can handle a 10 MHz sinewave from a signal generator.
• Watch out for unpredictable effects during bench testing. Some Gunnplexers may exhibit a swamping effect when aimed at each other in close proximity. If you suspect that desensing is preventing your link from working on the bench, try separating your transceiver units by at least a few feet, and/or place antistatic IC-storage foam in their apertures to act as an absorber. Although both Glenn Elmore and Alan Rutz have warned of this effect, I did not encounter it with the M/A-Com 10 GHz units, even with the 100 mW Gunn diodes installed. (Given the delicacy of the Schottky mixer diodes, it's probably not a good idea to aim the two Gunnplexers directly at each other on the bench with the antenna horns and high-power Gunn diodes installed.)
• Test the link thoroughly, and don't settle for 'twitchy' operation. Once everything is working properly, the link should be indistinguishable from a hardwired cable, when used to connect two relatively-quiet collision domains. If you observe flaky or inconsistent operation under any circumstances, something's wrong. The test suite I used to validate the prototypes included the Windows 95 "ping" command for light-duty, burst-type traffic; the Windows/DOS "dir" command operating simultaneously on mapped network drives at both ends for medium-duty traffic; and the Harris Semiconductor LANEVAL program for continuous one-way high-bandwidth testing near 10 megabits/second. LANEVAL was designed for testing Harris's 11-mbit/s Prism radio chipset, and until recently was available as a free download on the Harris Semiconductor site. The LANEVAL distribution file seems to have disappeared amidst a recent site reorganization, so I'll mirror it here until they put it back on their own site or yell at me to remove it. :-) Note that it's not uncommon for LANEVAL to indicate rates of only 5-6 megabits/second even on a hardwired 10BaseT connection. The exact figure achievable seems to depend on the network cards in use. Before blaming the microwave link for poor LANEVAL performance, repeat the test with a conventional 10BaseT or AUI crossover connection. 9.5 megabits/second is about the maximum real-world throughput for a "10-megabit" connection.
• Disconnect the transceivers from the Ethernet wire before taking down the link. A subtle pitfall associated with the microwave link is its tendency to spray random noise into the AUI receive-data line when a quieting signal from the other end of the link is not available, in a manner reminiscent of a TV set or FM radio tuned between stations. Some combinations of PCs, network cards, and operating systems seem to tolerate this random noise without objection. Most don't. It usually takes about 5 seconds' worth of this treatment to lock up a Windows 95 box hard. During the course of initial link testing, as well as in actual operation, you will want to be careful to unplug the AUI connection from either end of the link prior to powering down the transceiver at the other end, or taking any other action that would interrupt or block the microwave signal. As an improvement to the current design, it would be a good idea to add a rudimentary squelch circuit to the output buffer assembly which would suppress the transmission of noise into the Ethernet connection whenever the RSSI output from the MC13155 falls below 15 millivolts or so.

Component and Data Sources