13cm LNA design.
When I did my electronics degree, all those years ago, I wasn't really paying much attention to RF design, so I've had to learn a lot of this from scratch, as I failed to pay sufficient attention way back then.
I thought I would have a go at building an LNA for the 13cm band as I had a transverter that didn't really seem to perform as well as I wanted.
13cm/2320MHz is high enough to be interesting for design, yet should still be accessible with FR4 substrate for the circuit boards. Certainly, for higher bands, PTFE substrates are becoming more appropriate, but 13cm should still allow FR4, and as my initial attempts are likely to be needing fine tuning, FR4 is a much more "experimenter friendly" choice.
There are essentially 3 parts to the design and evaluation process:
The basic schematic with the bias circuitry.
The RF design
The basic schematic and bias scheme were as a result of my choice of active device for the LNA. I looked for something that was not obsolete, not too expensive and was not a "50 ohm in/out MMIC" as I wanted a bit of design fun with this. There is a lot of choice for this band, as it is widley used in the mobile phone and WiFi world. In the end I chose a device from MiniCircuits, the SAV541+ . A cheap PHEMT that should, in theory provide noise figures around the 0.5dB mark. This should be more than sufficient for terrestrial weak signal communications.
The basic schematic is shown below:
Power is provided via a LC low pass filter, a reverse-polarity protection diode and 5V regulator. The bias circuit is from the SAV541 datasheet and is a self-adjusting constant current scheme, which avoids the need for a current adjustment potetntiometer. Pots are a constant source of trouble in damp conditions, and best avoided. The rest of the circuit is very conventional for a JFET low noise amplifier.
The SAV541+ datasheet not only provides a full set of S Parameters in a touchstone .s2p format, but also some input matching parameters for optimum noise performance. For RF design, I used MatLab and a package called SimSmith that is an excellent (and free!) Smith Chart package.
My first attempt at noise matching was incorrect, more on which later, but I will describe he the second attempt, which I have a lot more faith in. Lets first look at the input match. From the datasheet, the following reflection coefficien data is found:
For best noise figure at 2.3GHz, I used a linear interpolation between the figures for 2.0 GHz and 2.4 GHz, settling on an optimum reflection coefficient of 0.36/121 degrees. This is the impedance that should be presented to the source port of the device to produce the optimum noise figure normalised to 50 ohms. Converting the reflection coefficient of 0.36/121 to a complex impedance gives us 29 + j20.6 Ohms. To establish the combination of lumped constant components and microstrip to match this impedance, I used SimSmith. I picked a suitable series capacitor and then experimented with lengths of stripline to produce a good match. I ended up picking a 32 ohm stripline, of 22.5 degrees and a 3.3pF capacitor.
Having established the input match, we now need to make sure the output of the device is well matched to 50 Ohms output when it is fed from that input.
It is possible to determine the optimum reflection coefficient for the output match from the formula:
Where ΓL and Γs are the load and source reflection coefficients and the asterisk indicates the complex conjugate. This was evaluated in Matlab to find ΓL as 0.97 at 74 degrees, which is pretty close to a "good match" anyway, so very little output match would be needed.
It is however possible to arrive at the same result through graphical means using SimSmith, which effectively performs the calculation above internally. We take the input circuit and present it to the source port of the device and then use SimSmith to produce a good match to 50 Ohms. SimSmith can take the S Parameters file for the device and produce a truly interactive match. If we were trying to match for optimum power gain, we would adjust both the input and output cirguits to attempt to see 50 Ohms looking into the input at the same time as we see a good 50 Ohm match at the output, but here, the input refelction coefficient has already been set from the datasheet, so we will restrict any adjustments to the output match.
The input matching components are copied over and an output match established, using the simplest arrangement of stripline and output capacitors possible for a resonable output match. The value of ΓL calculated above can be verified as the point at the right hand end of the turquoise line.
Having now established our input and output matching circuits, we can no move on to phyisical design and calculate the dimensions of the microstrip elements. The input microstrip is 22 degrees of 32 Ohm, the output stripline is 14 degrees of 75 Ohm, using a microstrip impedance calculator physical dimensions for these were found and the results used to generate the physical elements on the design.
1.2mm FR4 was chosen, the active device was placed as close as reasonably possible to the input socket, with 0603 size surface mount components ( 1.5mm x 0.8mm ) used for the majority. A 1K resistor was used for input bias in preference to an inductor, as this avoids problems with self-resonance.
Boards were ordered, solder paste applied and components hand-placed with tweezers before a quick trip through the desktop solder oven.
The units were then cased in the all too familiar tinplate boxes, with some reasonable SMA connectors soldered in place.
In order to evaluate the gain and noise figure, an Eaton 2075 noise analyser was used, with an HP 346B noise head. The 2075 only extends to 2.0 GHz, so while inidicative readings were taken at 2.0GHz for ease of use, a transverter was also used at 2.3 GHz as an external down converter, the calibration of the 2075 effectively removing the transverter from the measured figures. In practice, the figures measured at 2.0 GHz directly and at 2.3GHz with the transverter were within 0.1dB of each other, which is well within the level of uncertainty of the measurement.
replace image with correct LNA measurements.
In the first (incorrect) attempt, 1.6mm FR4 boards were used. The input pad was also incorrectly sized as 54 degrees of 33 Ohm. The resulting noise figure was 1.1dB at a gain of 16dB, gain was somewhat lower than hoped for and the noise figure somewhat higher. After re-reading the literature and gaining a better understanding of the techniques, I trimmed the input pads down to size with a Stanley knife, resulting in noise figures of around 0.76dB and 16.5dB of gain averaged across 3 test units.
The excess noise of ~0.25 dB is attributed to radiation losses due to the device being further away from the input than optimum, the use of 1.6mm board and the imperfect matching following manual modification of the boards. it seems reasonable that with a reduction in input losses through better placement of the device near the input, the use of thinner board and the correct input match, a figure closer to 0.5dB should be attainable.
New boards have been ordered in 1.2mm FR4, this will be updated when they have been assembled and measurements taken.