1 of 14 Balanced Amplifiers The single amplifier meets the specification for noise figure and again but fails to meet the return loss specification due to the large mis-matches on the input & outputs. To overcome this problem one solution is to use a balanced amplifier topography, which is shown in Figure 1. NOTE Dashed lines are reflected signals Figure 1 Schematic diagram of a balanced amplifier using two quadrature hybrids (eg Lange Couplers). The balanced amplifier employs two quadrature hybrids in this case two lange couplers (although branchline couplers can be used). Any reflections of an incident signal on the input due to the poor match of the amplifiers will be channelled back through the input lange to the 50 ohm load where they will be absorbed,and similarily on the the output.therefore if we look into the amplifier we will effectively see the 50 ohm loads and will therefore present a good match match.in addition this configuration will give us an extra 3dB s of output power (less the insertion loss of the Lange coupler ~typically < 0.25dB) and also the 1dB compression will be approximately 3dB s higher (In other words the circuit will be able to handle double the power without distortion).the main drawback of this circuit is the power required for two amplifiers instead of one. Amplifier linearity All amplifiers are designed to work over a given dynamic range where the amplifier should behave linearily, and generally with LNA s this is the case the input signals being received are very small.however there may be large signals out of band that may de-sensitize the LNA (ie reduce it s gain and therefore effectively increase the Noise figure of the overall system) or may generate intermodulation products that become mixed with other products further down the receiver in the mixer forming spurious responses.
2 of 14 The graph in Figure xx shows a typical 1dB compression characteristic for an amplifier. The Dynamic range is characterised by linear gain ie where the output power rises linearly with an input power. But as the 1-dB compression point is neared the output power begins to level out at the saturated output power level. At this point further increases in input power fail to raise the output power and in effect the gain has fallen to 0dB. Of a greater consequence to LNA design is the level of intermodulation products that are produced when two equal carriers are applied to an amplifier. The graph shows that the 3rd order intermodulation products rise at a rate of 3 to 1 with the input power. Output power Typ ~10dB Intercept Point Saturated output power 1dB compression point 1dB IM3 Input power Figure 2 Typical amplifier linearity plot showing how the output power eventually rolls off (saturated output power). Therefore, if we use a balanced amplifier the input power is equally spilt ie is 3dB lower therefore any IM3 products will be 9dB s lower and the 1dB compression point for the whole amplifier will be effectively 3dB s higher.
3 of 14 Lange Coupler Design Lange couplers consist of very narrow coupled lines of a quarter wavelength coupled in parallel to allow fringing on both sides of the line to contribute to the coupling. To increase the coupling it is necessary to use very narrow gaps and to still further increase the coupling bond wire interconnections are used. The resultant coupler will have a large bandwidth of at least an octave and so for our application we will have plenty of bandwidth with a design centred on 5GHz (Lange bandwidth ~ 4 to 8GHz). The initial analysis involves calculating the odd & even line impedances and then using a graph to read off the finger spacing and line widths:- Z oo / Z oe C R 1+ (c + 1)(N -1) 1+ 2 ( 1/ C 1)( N 1) Where N Number of coupled lines & C Coupling coefficient 2 10 db/20 Design for a 3dB coupler @ 5GHz using 4 coupled lines on Alumina ( εr 10.2) C 10-3/20 0.7079 0.7079 R 1+ (0.7079 + 1)(4-1) 1+ 2 ( 1/ 0.7079 1)( 4 1) 2 R 0.29786 Z on 50Ω Z oo. Z oe Z on [ N 1+ R]( [ N 1) R + 1] ( 1+ R) Z oo. Z oe 50. [ 4 1+ 0.29786]( [ 4 1) 0.29786 + 1] ( 1+ 0.29786) 96.272Ω Z oe ( Z. Z ) oo R oe 2 176.4Ω Z oo 2 ( Z. Z ). R 52.5Ω oo oe Using the calculated values of Zoo & Zoe we can use the graph of Figure xx to read off Values of S/d and W/d given that d will be 0.635mm.
4 of 14 Figure 3 Plot of Zoo againt Zoe The resulting values of S/d & W/d were found to be:-
5 of 14 S/d 0.1& W/d 0.07 therefore S 0.635x0.1 0.0635mm and W 0.07x0.635 0.044mm. The length of the coupled lines will be : - λair c f 3E8 0.06m 5E9 0.06 λ/4 (in microstrip) 4 ε eff where ε eff 6.9 0.06 Length of Lange coupled lines 4 6.9 5.7mm 50ohm lines W0.635 W0.044mm Optimised values in brackets L 5.7mm ( 5.9)mm S 0.064mm (0.054) mm
6 of 14 The data was entered into the Lange model on the CAD and analysed. It was found necessary to slightly increase the length to lower the frequency response and to narrow the spacing to increase the coupling such that there was slight over-coupling between the two output ports. The frequency response of the two output ports is shown in Figure 4 and the input return loss of the Lange is shown in Figure 5 Figure 4 Frequency response of Lange
7 of 14 Figure 5 Input return loss of Lange The amplifiers and Lange couplers were finally analysed to produce a full set of plots characterising the performance. The following page shows the final layout of the balanced amplifier. The amplifiers have been arranged to ensure that the bias can be applied from the walls of the enclosure. The FET s are grounded using VIA holes but could be mounted directly to a metal carrier using two separate substrates for the input & output circuits. To avoid thermal failure the carrier would have to be made of Kovar the match the construction of the FET. The shorted stubs are in fact RF shorted using very small (small electrical length) capacitors grounded using a via hole pad. In reality a longer open circuit stub could be used but would be another 5.7mm longer (ie one quarter wavelength), either of these two possibilities need to be done in order not to short the applied DC bias to ground.
Figure 6 Final Layout of the Balanced Amplifier Sheet 8 of 14
9 of 14 3 PredictedResults (I) Gain & Noise Figure response (passband)
(ii) Input & Output return loss response (passband) Sheet 10 of 14
(iii) Wideband Gain reponse Sheet 11 of 14
(iv) Wideband return loss response Sheet 12 of 14
13 of 14 4 Summary of results The table below shows a summary table comparing the required specification with the predicted analysed results from the CAD.In addition the power consumption and 1dB compression point are given. Parameter Specification Predicted Result Notes Gain 10 ± 1dB 10.1 ± 0.3dB Input return loss VSWR < 3 ie > 18dB > xx db Output return loss VSWR < 3 ie > 21dB > xx db Noise Figure < 1.8dB <1.5dB +ve Power consumption -ve Power Consumption 1dB Compression Point 5 Circuit performance - 2 * (10V * 11mA) 220mW - 2 * ( 10V *1mA) 20mW - ~ 17.5dB Total power consumption 240mW The circuit is designed to be made on Alumina with a dielectric of 10 and will be etched to produce the amplifier matching circuits and Lange couplers. There will be a tolerance in the production due to errors in scaling the circuit using photo reduction and undercutting of transmission lines during etching. At microwave frequencies placement of components can be critical and a component that is placed say 0.5 mm out of position will allow addition of another length of transmission line slightly de-tuning the circuit. Opening out the gaps on the Lange will reduce the coupling resulting in a under-coupling situation and hence cause an in balance in the amplifier degrading it s performance. What is considered the most difficult parameters to predict in an amplifier is the spread of the S-parameters from device to device and from batch to batch. A rule of thumb is that the S- parameters can vary by 5% from batch to batch. It is possible to use the CAD to calculate the optimisation yield of the amplifier by varying all the S-parameters and re-analysing the gain & noise responses. For flight applications active devices are supply as a batch with what are Lot acceptance Test samples and individual test results. These LAT samples are tested to ensure that the batch supplied meets with a procurement specification, which basically agrees with the manufacturers data sheet. For critical applications it is normal to characterise the LAT sample to produce a batch unique set of S- Parameters that can be used to analyse the circuit. This way the circuit can be tweaked on the CAD to ensure the amplifier will meet it s specification with the given batch of devices. With any manufactured item there will be variations in the dimensions of various items in our case the lengths of the Langes or stubs, which will effect the frequency response of the amplifier as will any variations in the dielectric constant of the Alumina. With all these variations the circuit may well have out of specification results when the circuit is built. No amount of yield analysis will ensure a compliant design and as a result some way of altering the circuit is needed once built. This is done by adding tuning pads to the layout,
14 of 14 using allow pieces of gold tape or foil to be bonded in a position, so that the amplifier can be tuned to be compliant at ambient. After a few iterations of tuning it is normal to find a tuning pattern that can be used on all amplifiers to give satisfactory performance assuming that all the active devices are from the same batch. In addition, to the production variations there are variations in the component tolerances in the bias circuit that will in turn cause variations in the drain current and drain voltage.