YET ANOTHER TRY TO MAKE A SUPER LOW NOISE PREAMP FOR 10 GHz

by Marko Cebokli S57UUU

EME conference Paris 98

ABSTRACT: After some disapointing results with SMA / microstrip preamps, I have decided to build an preamplifier with waveguide input circuit. This paper presents the design of this preamp, together with test results of the first prototype. I haven't yet achieved the hoped-for results, but decided to publish it anyway - for the pesimists as a warning that this may be a dead end, and for optimists as a stimulation for their own experiments.

1. EXISTING DESIGNS

When I was building the preamp for 10 GHz EME, with SMA input and microstrip matching circuit, I found it impossible to get closer than about 0.45 dB to the theoretical minimum noise figure of the transistor. I got about 1 dB system NF with NE 32684, which has 0.45dB optimum noise figure at 10 GHz. I tried two different designs, measured with two different low ENR noise sources, so I'm quite confident in the results. A survey of the last few years of DUBUS magazine confirmed my findings. All published designs [1],[2],[3] achieve noise figures of about 1dB, with similar transistors. The best reported value [4] is 0.68 dB with NE 329 (a 0.37 dB device), but that is probably pure (single stage) NF and the design was not published.
Since the best SAT-TV LNB's, where the microstrip is coupled directly to the waveguide, have about 0.7 dB system NF, it is safe to conclude that 0.3 dB above device NF is the 'sonic barrier' for microstrip, on 10 GHz. On 24 GHz, things will be even worse.

0.3 dB may not seem much, and indeed, when transistors have had 2 or 3 dB NF, it was negligible. But with a 0.4 dB device, adding 0.3 dB of loss means almost doubling your noise!

Theoretically, considering the second stage noise contribution, receivers with 0.5 ... 0.6 dB system noise should be possible with these transistors in a low loss circuit, so I decided to try and make one.

2. DO WE REALLY NEED SUCH LOW NF's?

The opinion can often be heard, that on 10 GHz EME, because you can hear Moon noise, you are Moon noise limited, and further reduction of receiver noise makes no sense. But, in most cases, this is not true.
Let's consider a typical 10 GHz station: 3m dish, 80K (1.06dB) receiver, 40K antenna noise. A 3m dish will gather about 50K of Moon noise, so Moon / cold sky will be 170K/120K or 1.5 dB of Moon noise, a typical value for such a station.
That means that out of 170K of total noise, only 50K, or 29% is Moon noise - quite some room for improvement! A 40K (0.56dB NF) receiver would mean 130 K total noise (including the Moon), a 1.1dB improvement in total sensitivity. Moon noise in this case would be 130K/80K or 2.1 dB.
So where is the limit, when it really doesn't pay to improve the receiver any more? My suggestion is 3dB of Moon noise - that is, when half of your total noise comes from the Moon. In that case, you are still 3dB less sensitive than the (nonexistent) ideal noiseless receiver, but the improvements you can make become smaller and smaller and cost more and more.

If we someday get a satellite with a 10 GHz downlink (phase 3D), this will be another use for super low noise preamps. The absence of Moon noise will make them even more attractive, to have the smallest possible antenna.

3. CAUSES OF LOSS IN A MICROSTRIP PREAMP, AND SOME CONVENTIONAL CURES

In a SMA / microstrip preamp, there are several places where losses occur:

SMA connector. This can be most easily eliminated, as is the case in SAT-TV LNB's, where a probe goes directly from microstrip into the waveguide.
Coax / microstrip transition. Without a special 'microstrip launcher' type SMA connector, it is very hard to get a decent match. The solder bump also causes radiation losses.
Microstrip line itself. 0.15dB / cm is typical, and it is hard to make an amplifier with less than one cm of input line. Thick substrates usually have less ohmic losses than thin substrates, but more radiation losses than thin ones.
Input capacitor. Not so much absorption loss, but the cap makes a radiating bump on the line. Because most preamp inputs are connected to a coax / waveguide transition, this capacitor can usually be left away. Only noise measurement is problematic then, since noise sources have resistive attenuators at their outputs, that will short the DC. A good DC block for measuring purposes can be made of two SMA / waveguide transitions, screwed together back to back. I have measured the loss of such a combination on a HP 8510, and it was 0.15 dB. (0.075 dB each transition)
Bias circuits. Their loss is mostly radiative. G3WDG told me that he spent much time optimising these in his preamp.

In the above list, many losses are radiative, and some improvement could be made if we could prevent these things from radiating. Here we can learn another thing from SAT-TV LNB's. If you open one of these, you will probably see that the preamp is mounted inside a separate very narrow metal 'tunnel'. The trick is that if you put something inside a waveguide that is subcritical at a given frequency, it won't be able to radiate at that frequency, because the energy simply cannot propagate away. This also eliminates the problem of instability due to radiative interstage coupling, that people usually try to cure by inserting absorptive material. But it is BAD practice to put absorbers into a low noise gadget.

4. WHAT ALTERNATIVES DO WE HAVE?

I looked through lots of literature (IEEE periodicals, microwave conference proceedings etc) to see what people have done in this respect. Several people found microstrip inadequate for low noise amplifiers, and proposed a variety of alternatives, like coplanar lines or suspended substrate striplines. Two of them [5],[6] proposed mounting the transistor directly into waveguide. Marti [5] has mounted the transistor on a vertical (E-plane) substrate in a waveguide and made the matching structure using asymetrical finlines. Fig 1. (7kB GIF) Author achieved 2.2 dB NF on the 11.7 - 12.4 GHz band, with NE 710, a 1.6 dB device. But it is not clear from the paper, if this includes a bandpass filter in front of the amp. The weak points of this design are the gate bypass capacitors and the problem of assuring ground continuity across the substrate.

Angelov et al. [6] propose a variety of low noise structures. For a 22GHz LNA they have coupled the transistor to the waveguide without using any dielectric substrate. Fig 2. (7kB GIF) They have achieved 2.3dB NF with MGF1403. 21GHz is above manufacturer's data for this transistor, by extrapolation we can estimate its minimum NF on 21 GHz to be about 2.4dB. This is probably an error in extrapolation, but it could also mean that the test fixture that the manufacturer uses to measure noise is lossier than this waveguide structure.

5. MY DESIGN

I have decided to make the input matching according to [6], and to use normal microstrip circuit on the output, where losses are not so critical. The preamp consists of three main mechanical parts: the base Fig 3a. (44kB GIF) , the wavegude transformer Fig 3b. (42kB GIF) and the transistor clamp Fig 3c. (6kB GIF) . These parts are machined brass with silver plating.
The gate bias is supplied through a coaxial low pass filter type choke. The choke consists of two turned pieces of brass as shown in Fig 3d. (5kB GIF) They are linked by soldering to a piece of 0.5mm copper wire, and inserted in a hole in the base part. A 25 micron (1 mil) mylar film is wrapped around for insulation. A piece of 0.1mm copper foil, cut as shown in Fig 5. (6kB GIF) links the top end of the bias choke to the gate terminal of the transistor and serves also as the RF coupling loop. Its shape is optimised for a good impedance match. This match is quite sensitive to the shape and position of this piece of copper foil!
The transistor is mechanically fixed by the source leads that are held in place by the transistor clamp. The drain terminal is soldered directly to a small microstrip circuit board. This board provides drain matching and bias. A SMA female flanged connector serves as the RF output.
The whole assembly is shown in Fig 4. (13kB GIF)
And here are two photos: Photo1. (27kB JPEG) Photo2. (34kB JPEG)
The input is a stepped height waveguide impedance transformer [7] that lowers the impedance to a value more suitable for transistor matching.
My idea was to make a good impedance match with this transformer and the shape and position of the gate coupling loop, and then shift the impedance with tuning screws in the waveguide towards the optimum noise point.
With modern (1997) HEMT's, the optimum impedance match and optimum noise points aren't very far apart, so this should be possible without introducing significant additional losses. (i.e. without putting the screws too deep into the waveguide)

6. MEASURED RESULTS

As of the time of writing, I haven't yet achieved better than 0.87 dB NF with a NE 326. That is about 0.13 dB better than my microstrip version with the same transistor, but still short of the desired results.
I have measured the return loss at the input with the transistor removed, and it was well below 0.1 dB (such values are hard to measure even with a HP 8510) So, it seems, the input circuit losses of less than 0.05 dB are not the problem.
The next possibility was that I'm shooting in the wrong direction (impedance-wise) with the tuning screws. But there is one well defined impedance point, namely that of matched impedance, which can be easily identified on a network analyser. Even at that point, the measured NF was significantly higher than the calculated value (circles of constant NF etc.)

For now, I'm trying to devise some more measurements to find the missing tenths of dB's. However, I'm also beginning to suspect the transistors. 5 to 10 years ago it seemed that the big volume users (Sat TV LNB makers) were buying all of the first-class devices that the device manufacturers could produce. The DEVICE NF's of the best transistors you could buy were equal to the SYSTEM NF's of the best LNB's you could buy at the same time! Also, the small volume resellers usually don't have the strict input quality control that the big volume users do. They are often hunting the bargain, which seldom takes the form of the first class merchandise. In Italy, even transistor label falsification was popular some time!

7. SOME NOTES ON NOISE FIGURE MEASUREMENT

Today, noise performance of LNA's is almost exclusively measured on microprocessor controlled PANFI's like HP7890B. These instruments tend to have a bewildering multitude of measurement modes, calibration options, special functions and so on. So, especially if you don't use the same instrument every day, it tends to be very confusing. That's why I have developed a 'standard procedure' for measuring noise figures on a HP7890.

It goes like this:

Let the NF meter warm up at least for one hour, with noise source connected.
Enter your room temperature under special function 6.0 (Tcold)
if the frequency is above HP7890B's range (1.6 GHz), I use for downconversion the transverter that I normally use for QSO's, complete with the same cable between the LNA and transverter that I use in real life.
Despite external mixing, I use mode 0, input frequency 144 MHz, and type in the ENR value of the noise source at the input frequency (10 GHz) as special function 5.3 ('use spot ENR')
I DO NOT use calibration, to IMPROVE accuracy!!!
If there are losses between noise source and DUT, I dont use the special function 34.2 for that, but just type in the ENR of the source minus the losses. For example, if the source has 5.55 dB ENR and I have a 0.075 dB coax/wg transition, I type 'use spot ENR', 5.47 .

This may seem odd, but here is the explanation:

Why no calibration? There is a big difference between calibration on a network analyzer like HP8510 and calibration on a noise figure meter. On a network analyzer, calibration removes lots of systematic errors (remember the name '12 term error model'), and can improve accuracy for orders of magnitude. In contrast to that, calibration on a noise figure meter only enables you to simultaneously measure gain, and to subtract the second stage noise contribution. If you made a bad calibration, or because of thermal drift, the calibration on a NF meter can actually DEGRADE your accuracy! Also, because one is usually interested in his total SYSTEM noise figure, it makes sense NOT to subtract the noise contribution of later stages. If these later stages are your real life transverter, by not subtracting its noise contribution, you will actually measure your system noise figure, and tune the preamp to the optimum compromise between noise and gain that minimises your total system noise! Of course, if you want to measure the 'pure' noise figure of your preamp, like on a NF contest, or want to characterize a device for noise / gain, you MUST use calibration. In that case, use lots of averaging during calibration, and measure your DUT as soon as possible after calibration. After measurement re-check the calibration! It can drift away a few tenths of a db in ten minutes!
Why not use the appropriate measuring mode for this type of external mixing? Well, just for simplicity, and to reduce the chance for errors (like forgetting to tell the machine that you are using SSB mixing etc). The purpose of these special modes is to tell the machine the input and IF frequencies. It needs to know them only to control an external variable LO, that we don't have, and to know which ENR to use from the table. Since we are measuring only on a single frequency, it is simpler to use spot ENR.

8. CONCLUSIONS

It still does make sense to go for the lowest possible NF. Since conventional designs fail to deliver what is theoretically possible, I scanned the literature for alternative designs, and made a waveguide input LNA. The improvement was measurable, but still not what I was shooting for. So, I plan to continue work on this design. The measurement of such low NF's means squeezing the last out of today's PANFI's, so I have described a measurement procedure that tries to simplify things and reduce the possibility of errors.

9. REFERENCES

[1] C & P Suckling G3WDG/G4KGC: Modern 10GHz Transverter System (Part II)
    DUBUS 4/93   pp 21-46

[2] Silvano Ricci I0LVA: HEMT Preamplifier for 10GHz. DUBUS 3/94    pp 52-57

[3] Michael Kuhne DB6NT: LNA for 10GHz. DUBUS 3/95    pp 5-10

[4] Microwaves USA: the '95 Microwave Update, noise figure contest results
    DUBUS 4/95
    
[5] Albert Marti: Ridge Guide Preamplifier for DBS Outdoor Unit;
    European Microwave Conference 1983 proceedings pp 221-225

[6] I.Angelov, A.Spasov, A.Stoev, L.Urshev: Investigation of Some Guiding   
    Structures for Low Noise FET Amplifiers; European Microwave Conference
    1985 proceedings pp 535-540

[7] J.Uher, J.Bornemann, U.Rosenberg: Waveguide Components for Antenna Feed
    Systems: Theory and CAD.   Artech House 1993.

Up to S57UUU EME Page