1.3GHz HEMT Tropo Preamp

Chris Bartram GW4DGU

Getting back on the bands after a long period QRT has been a far more complex, and time consuming business than I'd expected! Having put together 'legal limit' systems for 144, 432, and something reasonable for 10368MHz, I'm now starting to work on 1296. I like to design and build my own kit, even though it might seem a bit of a busman's holiday to some. I'm very fortunate to have excellent computing and lab. facilities as a consequence of my work as a freelance RF/microwave circuit designer, which makes life a bit easier! This preamp has been designed to be an excellent tropo preamp, and a good second stage for EME. In practice, a pair of the amplifiers cascaded would probably be just about good enough for EME, but some redesign could probably shave 10K or more off of the noise temperature, at the expense of simplicity and possibly unconditional stability.

Please don't expect this to be a detailed constructional article. It's not. I'm trying to informally describe aspects of a project I've completed for my own entertainment, and to to possibly provide a small amount of inspiration for others. Experienced microwave equipment builders should find enough information here to duplicate the design, but I hardly have enough time to earn a living, look after a small farm, and to play radio, let alone properly support a constructional project! If somebody wants to take the design further, and make PCBs available, I'd be be happy to help, though.

I set-out to design a preamp covering the whole 1240 - 1315MHz range which would be easy to make, with good, but not necessarily spectacular noise figure, adequate gain, unconditional stability and good linearity and output return loss.

The bandwidth of the preamp is very large. In gain terms, although not in terms of noise figure, it's still usable at 432! As it has good intermodulation performance, in many locations the amplifier could be attached directly to a 1296MHz antenna, and not suffer from intermodulation and out of band issues. Living in the primary service areas of two major TV broadcast transmitters, I can use the amplifier connected directly to a 1296MHz yagi antenna without noticing intermodulation products, but I may be lucky!

In more stringent situations, a low-loss bandpass (or highpass) filter could be connected to the input. From a systems viewpoint, it makes good sense to separate the functions, rather than trying to design a narrowband preamp. If the filter and antenna have good return-loss at 1296MHz, the amplifier will still see something close to 50Ω, so the noise match will be uneffected, and the noise figure of the filter-amplifier combination will be the sum of the preamp. noise figure and filter loss. I'm working on the design of a simple-to-make, but very low loss bandpass filter, which I'll be putting ahead of this amplifier as a precaution. There are a number of options for a post-amplifier filter. My 'final' solution will probably be to use a second low-loss input filter as an interstage filter, as my planned new transverter will have an excellent narrowband response from dielectric resonator filters.

Filters and antennas present impedances which can be far removed from 50Ω outside their passbands. For an amplifier to be stable with any combination of passive input and output load over the range of frequencies where the active device has gain is a highly desirable, but often difficult requirement to meet. It's even more difficult to prove in the real world. This preamplifier has been designed to be stable using the usual stability measures, and a detailed model been simulated to beyond 15GHz. It shows none of the usual signs of instability in simulation, and I've so far failed to see any on the bench. That doesn't mean that there isn't a frequency somewhere in the spectrum where a combination of passive source and load impedances couldn't provoke instability. It just means that I've not yet found it!

Performance

My prototype achieved the following performance at 1296MHz:

Noise figure: 0.55dB (T ≅ 40K)

Gain: 12.1dB

Input 3rd order intercept: -4dBm

Input 1dB gain compression: -11dBm

Output return loss: 20dB

The measured data agrees well with my simulations, and also meets Bartram's First Law of LNA design: in the absence of linearisation circuitry, the input third-order intercept of any low noise device, biased for low-noise operation, will be of the order of 0dBm.

The plots were obtained from my VNA and written to a file via the IEE488 bus, and then plotted using OpenOfficeCalc running under Linux. The return loss graph looks a little noisy, as I accidentally read the data at 1dB resolution, and didn't notice until I came to plot it.

My simulations also suggest that the noise figure remains good across the 1.3GHz amateur band, and is still OK at 1420MHz, I'd expect that to be the case, in practice, but I haven't yet measured the amplifier at other frequencies.

Although apparently relatively simple, noise figure is a difficult parameter to measure accurately, particularly with modern test equipment! I am sceptical of many claimed noise figures. My measurement was made using a professional semiconductor noise source, and my spectrum analyser, preceeded by a low-noise broadband amplifier. I guesstimate that the accuracy of my measurement is within about 0.3dB. A more accurate - and probably appropriate - method for amateur measurement of modern LNAs would be to return to the basic physics, and use a low thermal mass 50ohm termination dipped altenatively in melting ice and boiling water. However, that doesn't quite have the cachet of a £30k item of test equipment, and it requires a modicum of understanding rather than the ability to read a display uncritically...!

Background

I have a number of Fujitsu FHX05 HEMTs in my component drawers following a successful Ebay bid(!). OK, there are better devices, but not much better, and at £1 (€1.50) each.....?! Using Fujitsu's published device models, it was clear, following a lot of analysis, (using the Eagleware Genesys software I use in my work) that it it wouldn't be possible to guarantee unconditional stability out-of-band using a 'source feedback' topology. As an important aside, although the source feedback topology has been around for twenty years or more, and is highly trendy - probably because it's possible to obtain a reasonable input match without degrading the noise figure - I've never found it possible to make a completely stable amplifier when using it. It's nearly always possible to find a region of instability, often at tens of GHz..! I know of at least one other UK 1.3GHz preamp project (using Agilent PHEMTs with source feedback) which foundered for exactly that reason. For a LNA, a good input match isn't actually necessary. Good input matching won't provide any more sensitivity (it's a matter of getting the right 'mismatch') but instabilities can completely wreck a potentially good NF. I've seen suggestions that a good input match can help with stability, but that's not true - although with a marginally stable amplifier it can sometimes seem so. Neither will reflection losses due to filters placed between the preamp and antenna add to the noise figure. Provided the amplifier sees a source impedance of close to 50ohms, the noise figure will be close to the sum of the resistive filter losses (in dB) and the amplifier noise figure (corrected for subsequent stage contributions).

The design that emerged after a number of simulations employs a conventional mismatched input circuit realised using lumped inductors for the input impedance transformation and gate bias feed and a shunt capacitance formed by a short length of microstripline. This forms a shunt-L, series-L, shunt-C network, which allows gate bias to be introduced at a relatively insensitive circuit node.

The HEMT source connections are grounded via 1mm diameter pins cut and filed flush to the top (component side) surface of the PCB. This is critical for stability.

The other component grounds are made by wrapping a piece of copper foil around the edge of the PCB at the appropriate places.

The output circuit is broadband. In order to control potential instabilities, caused effectively by the output resistance of the FET going negative at some frequencies, a small series resistor has to be inserted in the drain circuit. This causes a small degradation in the obtainable noise temperature, but not as much as the device hooting away merrily at 30GHz! There is also a gain penalty, but gain is cheap, nowadays! The drain load is an inductor-resistor series network. This gives a good broadband output match.

PCB

My prototype PCB was cut by hand from 0.75mm (0.030 inch) thick Rogers R4350 pcb material. I used just a scapel, and a steel rule, working under a X10 binocular microscope. Using this technique I've made prototype LNAs to past 10GHz, and power amplifier boards to several GHz. R4350 is a low-cost, low-loss PCB material intended for large quantity RF/microwave applications. It can apparently be processed by using standard production techniques employed for FR4 material. It's probably not part of the manufacturers design remit, but R4350 is also particularly easy to work by hand! I have generated a set of Gerber files for the PCB layout and they can be found on this website. These may be used for non-commercial applications.

The drawing here has been through a number of format translations, and is adequate as an illustration. It isn't to scale, and a number of 'funnies' have crept in. I'd recommend that even if you intend to adopt the scalpel approach to PCB production, you download a Gerber viewer, such as PREVUE and use that to print scale drawings of the PCB.

Components

The most critical components are in the input circuit. The 6p8 capacitor really needs to be a low-loss part. I used an 0603 AVX 'Accu-P' capacitor. Porcelain capacitors could be used, and even, at a pinch, a standard 0603 COG cap although that would have an effect on the obtainable NF. The 18nH shunt inductor is reasonably critical. A low-Q device could degrade the noise figure by some tenths of a dB. My choice was a Coilcraft 0805CS series wound inductor.

Perhaps the most critical part is the handwound inductor. This is wound with a constant 2.5mm pitch on a 4mm mandrel eg. a 4mm drill shank, such that the spacing of the centre of the start and finish of the winding is 5mm. There are no leads. Simply trim the coil so that it is a two turn helix. I used silver plated copper wire because I had some! Enamelled copper wire would also be suitable. In practice, unless high conductivity silver plating is used - and it's protected from corrosion - there's little advantage over copper. Don't even think of using tinned copper wire....

It may be necessary to slightly squeeze or stretch the inductor to optimise the noise figure, depending on exactly how the amplifier is built. If you feel this is necessary, thoroughly ground the gate of the HEMT before you do anything, and remove the solder at one end of the coil. Make your modification, resolder, and then remove the gate grounding. Otherwise you'll either lift a pcb track or damage the HEMT, or both! I know!

All other passive components were standard 0603 parts. The capacitors should have COG dielectric. In my case, the 5n6 inductor was a Coilcraft 0603CS wound part, but monolithic inductors would also be suitable.

The FHX05 is the middle device in a series. It looks from the data sheet as though the devices are selected for noise figure in the 11GHz satellite television band. The FHX04 has better guaranteed NF at that frequency, and the FHX 06, slightly worse. I suspect that there would be very little difference between them at 1.3GHz.

Enclosure

Although it might just be sensible to use a milled enclosure for a preamplifier built on a flexible substrate like ptfe/glass, that's really overkill. I tend to solder preamp pcbs into a lidless brass shim 'case'. This is more for physical protection than for RF screening, as I don't treat screening as a universal prophylactic! In this case, the bare PCB is small, the substrate material is relatively rigid, and providing the board isn't maltreated by being flexed, the amplifier will work entirely adequately just hung in the wiring! Flexing will break surface-mount components very easily, and even with a microscope, it's sometimes difficult to detect this visually.

Although there are, of course, many situations where good screening is mandatory, it can bring its own problems. This is particularly true of amplifiers using devices with bandwidths of tens of GHz. There are hazards in packaging microwave amplifiers: putting-on lids leads to many potential problems as it is frighteningly easy to excite cavity resonances. Identifing and killing these can be as much work as designing the amplifier in the first place!

HEMT precautions

It's very easy to damage HEMTs and not realise it. They are extremely susceptible to static damage during assembly, and from supply line transients in operation. This doesn't usually show as a significant change in the dc parameters, just as a change (for the worse!) in noise figure.. Be VERY careful, and take extreme precautions to avoid static damage when soldering the device into circuit.

Power supplies

The preamp requires a drain supply of +2.8V, 10mA. The gate bias required for this will be of the order of -0.6V.

As HEMTs, along with most LNA devices are susceptible to supply-line transients, I now operate my preamps from isolated power supplies. In this I use a cheap packaged transformer-coupled inverter which converts 8 - 36V dc into ±5V, this largely isolates the preamp from supply line transients, and lets me power the preamp from my 28V antenna relay supply, as I energise the antenna c/o relays on receive. Following the inverter, I filter and clamp the ±5V lines and use these to power active bias networks. I'll write-up this 'preamp power supply for the paranoid' in the near future, along with details of the sequencing circuitry I use.