Experience with Power Amplifiers
Chris's experience with (thermionic and solid state) power amplifiers is quite extensive and diverse. He has personally installed, commissioned and operated travelling wave tube and klystron high power amplifiers in redundant configurations (about 200 W) on satellite ground stations and on point-to-point tropospheric scatter stations. He has also been charged of the technical content of proposals for satellite and terrestrial ground communications stations which included travelling wave tube amplifiers (TWTAs) with the associated redundancy and switching arrangements. RF interconnections to these were both coaxial cables and rectangular waveguides. He is familiar with safety measures required with high voltage power supplies and the risks due to exposure to high flux densities of non-ionising radiation.
He has supported the design of communication satellite payloads including TWTAs and solid state power amplifiers (SSPAs) based on combined GaAs MESFET devices at C, Ku and Ka band. This work included developing various 'one for N' type redundancy configurations and looking at using multiple port amplifier architectures as an alternative to simple switched redundancy.
At intermediate powers (a few 10s of watts) covering HF and VHF he has designed and successfully built various class A and class AB push-pull amplifiers using MOSFET devices at HF and the Philips BLU56/86 bipolar devices at VHF. These projects started with full small signal S-parameter characterisation for the Rollet stability calculations and used datasheet-provided provided high power source and load conjugate match impedance data for input and output (simultaneous conjugate) matching and full thermal and biasing design. He has a good understanding of power amplifiers in general including single ended, push-pull, their classes of operation (Pozar, Gonzalez, Dye & Granberg at Motorola) and ways of increasing their bandwidth. His good all round experience of RF engineering has proved to be invaluable in appreciating the particular challenges of of high power RF design.
He has developed wideband balun transformers using soft ferrite materials based on work by Sevick and gone on to use them in class B/AB push-pull power amplifers at VHF, both FET and bipolar designs. The ferrite materials used were F9, F14, F19, F52 (MMG Neosid) and K1, and U17 (EPCOS/Siemens). The successful designs achieved close to theoretical splitting loss and around 30 dB wideband isolation at HF and VHF. They operated well at intermediate powers with minimal absorption and it was expected that they would handle much higher powers easily.
RF Power Amplifiers
RF power amplifiers (PAs) are those often used in the final stages of RF transmitters in order to generate appreciable output power, sufficient for transmission, often into an antenna. Single transistor or single-ended PAs can typically deliver output powers up to 10s of watts at VHF, but this reduces as the frequency increases. The outputs of 2 or more transistors may be coherently combined to obtain higher output powers. There are several advantages from designing PAs in modular form, each PA comprising 2 matched stages in parallel. PA stages, compared to other stages, usually consume the greatest DC power and therefore usually represent significant battery drain in portable and mobile equipment. It is common practice therefore to operate PAs slightly into compression, typically to around the 1dB compression point (P1dB), in order to achieve greater efficiency, provided the extra distortion products can be tolerated. Linearisation techniques may be used to reduce the distortion, but normally at the expense of a degree of efficiency.
S-parameters are nearly always defined for linear conditions at small signal levels and are therefore suitable for use at low level stages which usually operate in linear mode. This type of S-parameter data are not suitable for use in designing PAs for operation into non-linearity, since the parameters will tend to change when the operating point moves away from linearity. An exception is where a PA is designed for class A operation since, by definition, this will operate only in the linear mode. Tests should be conducted to confirm that it will in fact normally operate in linear mode.
Devices that are designed for PA applications will normally include information on how they perform at high power levels, both electrically and thermally. Typically these will provide data on the input and output impedances of the device for conjugate match to the system impedance at the rated power level (often the P1dB point) as well as the associated DC bias settings (IC and VCE for a bipolar device or IDS and VDS for a FET device) and temperature. This may be provided as rectangular graphs against frequency or as two families of plots on Smith charts. For each test frequency, such Smith chart data normally comprises a set of roughly circular contours, known as power contours, each one for a particular output power level, the levels increasing towards their center. Generally, for bipolar devices, the impedances will be much lower than 50 Ω and therefore towards the left of the (normalised impedance) Smith chart. Another useful parameter is the compressed gain, again usually for the P1dB operating point. This may be tabulated or expressed as a graph of CW gain against the input power at a defined test frequency. This would be necessary, for example, to determine the required drive level into the PA for a given output power to an antenna.
The power contours provide the normalised input and output impedances or reflection coefficients (normally looking into the device) for various output powers close to and including the maximum, or the information could be supplied in tabular form. Once the power operating point is chosen, suitable low loss matching must then be designed such that the conjugates of these reflection coefficients are seen looking into the source match and looking into the load match.
Practical Design Issues with Power Amplifiers
High powers mean high level RF currents flowing through the connections of the RF components, particularly near transformer and combiner grounds and tracks close to the active devices. Connections must be low inductance and low RF resistance, taking due account of the associated skin depth. Wide ‘ribbon type’ conductors and high Q capacitors must be used with connections of the minimum possible lengths. Grounds planes must be used extensively with multiple ground parallel through holes as close to the devices as possible. For higher powers improved grounding using heavy copper conductors may be required.
The first step in considering a device for use in a PA is to calculate the Rollet stability factor across a frequency range well in excess of the intended operating frequency range. Quite often an RF PA will be prone to oscillation at a much lower frequency than the RF frequency range it is designed for as here the device gains tend to be greater and decoupling capacitors have higher reactance. As the onset of instability usually starts at low signal levels, conventional small signal S-parameters may be used for this, measured over a suitably wide frequency range. With small signal amplifiers, an unintended oscillation although very annoying, does not usually cause any damage. With PAs however, such an oscillation can occur suddenly and dissipate enough power to destroy the amplifying devices and the designer must be reasonably sure that no instability will occur at the first power-up. The device does not have to necessarily be unconditionally stable across the full operating frequency range because the necessary power conjugate matching might bring it into the stable region. However caution is required.
(example of device small and large signal S-parameters, see Pozar p 571).
Single Ended Configurations
Single ended PA configurations may be realised in lumped elements, distributed components or a combination of both, depending on factors such as the specified frequency and bandwidth, the space available and the operating power level. They are typically suitable for frequencies up to a few hundred megahertz and power levels up to about 10W or lower powers into the microwave range where distributed components would normally be necessary. Relatively narrow band designs can be realised with LC tuned circuits. Wider band designs may use input and output coupling transformers.
Advantages: Cheap and straightforward (no balun transformers are required).
Disadvantages: The spread in transistor characteristics must be tuned out individually using adjustable capacitors or adjustable inductors. Difficult to make broadband.
Classes of Device Operation
There a 3 basic classes of operating an amplifier device: class A, class B or class C.
An amplifier will have some form of general dynamic transfer characteristic which will exhibit 3 regions:
i) the threshold region
ii) the linear region
iii) the saturation region
The operating (DC bias) point for class A is approximately half way along the linear region between threshold and saturation. Class A is the most linear mode of operation in that the output waveform is an accurate replica of the input waveform and a change of 1 dB in the input signal will cause an identical change of 1 dB in the output signal. The output waveform is present for 360° of every input cycle or the conduction angle is 360°. This class of operation only remains linear provided that the input power is such that the peak-to-peak amplitude of the input voltage waveform does not extend beyond the linear part of the dynamic transfer characteristic.
If the input signal level is zero, a DC bias remains and the device still dissipates power. If the active device is a bipolar transistor, this is known as collector dissipation and will be dissipated directly into the collector heatsink which must therefore be designed with sufficient thermal capacity to absorb it. Class A is relatively inefficient, the maximum theoretical efficiency being only 50%, requiring appreciable space to be necessary for heatsinks and power supplies. As the input signal level is increased from zero, the collector dissipation will reduce and the signal output power increase. Therefore the efficiency will increase from zero for zero input signal level to approaching 50% when the output signal voltage swing occupies just about all of the linear region. Class A is ordinarily used for low power stages where the minimum distortion is required and inefficiency is not an issue. At later stages, as the signal level increases and DC drain becomes more significant, a class A stage can take significant DC power and, in the case of portable equipment, drain the batteries quickly. As the input signal level is increased further the instantaneous voltage peaks will eventually go into the non-linear regions of the characteristic and the amplifier will start to compress, causing distortion and no longer operate as class A.
Class A is therefore used for high power stages only if high linearity is important and there is sufficient capacity in the power supply. For example in mains-powered test equipment and add-on linear amplifiers.
In class B, the DC bias point is set to the threshold value of the dynamic transfer characteristic which is also known as the cutoff point. Therefore the conduction angle is nominally 180° and the output is present for alternate positive half cycles of the input waveform. (In fact if there is significant cross-over distortion, the conduction might be slightly less that 180° - see below). In most cases, where 360° amplification is required, two parallel class B stages are required operating in anti-phase, so that each side operates on alternate half-cycles of the input waveform. This type of operation is known as push-pull and may be achieved either with a hard-wired complimentary configuration using matched NPN and PNP transistors or alternatively using balun transformers as described below. Few such NPN/PNP transistor matched pairs are available for frequencies above audio.
The collector dissipation is virtually zero with no signal applied and gradually increases as the input signal is increased. The power supply must have sufficient capacity and sufficiently low impedance for the maximum output signal level. The maximum theoretical efficiency is relatively high at 25π% (78.5%) and this class is often used for battery powered applications,
Push-Pull Using Balanced/Unbalanced Transformers
Balanced/unbalanced transformer (balun) coupled push-pull operation is commonly used up to UHF. At higher frequencies it is more difficult to get an accurate 180° phase shift across appreciable frequency ranges and conventional transformers become very lossy so push-pull is not normally used except for very narrow band operation. Assuming two similar devices, the advantages of push-pull compared to the single ended configuration are:
The push-pull circuits must be designed with close symmetry. The push-pull requires a balun transformer (unbalanced at the input to balanced at the output) at the input and a similar one but connected in reverse at the output. Designs may be with distributed components at higher frequencies and lumped components at lower frequencies, or a combination of both.
The center tap of the balanced side of the balun does not necessarily have to be grounded. If it is not, there will be some tolerance of imbalance between sides. The signal return path for the 'on' transistor is via the input capacitance of the 'off' transistor. The balun can be designed to provide some of the impedance matching by appropriate choice of turns-ratio (lower frequency flux linkage type) or the arrangement of transmission-line impedances at higher frequencies.
In class B the input signal level at the transistor must be sufficient to overcome the base-emitter forward biassed voltage drop before the device starts conducting and generating an output. This is about 0.6 V for silicon devices or 0.2 V for germanium devices. The absence of output for the very low instantaneous input voltages causes distortion of the output waveform of the type shown below, known as cross-over distortion. Cross-over distortion may be eliminated by operating the device in class AB instead of class B.
Class AB is very similar to class B, but instead of no bias, a small bias voltage is applied to just overcome the base-emitter forward biassed voltage drop, of approximately 0.6 V for silicon devices or 0.2 V for germanium devices. In this way the amplification is linear also for small signals and cross-over distortion is virtually eliminated.
In the class C configuration the DC bias voltage is set to a value below cutoff. The device will only conduct when the sum of the bias voltage and the instantaneous input signal voltage rises above the threshold value. Therefore the conduction angle will be less than 180°, typically being set to about 140°. The output signal will be a very distorted version of the input signal but with significant content at the fundamental frequency. The output coupling network of a class C stage requires a sufficiently narrow band filter to reject harmonics of the applied signal but sufficiently wide to pass the modulation and frequency variation of the input signal. The efficiency of class C can be high up to 80% or 90% but the limited percentage bandwidth makes it most suited to RF.
Matching for Power Amplifiers
Bipolar devices that are intended for use in power amplifiers normally have test data defining the input and output impedances at rated power levels as a function of frequency, when conjugately matched into the system impedance, usually 50Ω. In general these will differ from the impedances derived from the S-parameters which are normally measured for linear conditions at low signal levels. The matching networks used to generate this data require to be of the lowest possible loss and of sufficiently high Q to reject the harmonics adequately. Power device packages are designed with high thermally conductive electrodes connected to flanges or studs that may be used to dissipate power. The maximum junction temperature for a silicon bipolar transistor is about 200°C and the maximum channel temperature for a GaAs FET is about 175°C.
Constant Output Power Contours
AM to PM Conversion
AM to PM conversion is a phase distortion especially relevant to power amplifiers which carry digital modulation such as QAM and results in a change of phase through a power amplifier as a function of the instantaneous amplitude of the input signal. Such changes increase the probability of symbol errors.
Parallel Combining of RF Transistors
Some transistors may be parallel combined to achieve greater output power than may be obtained with a single device. Paralleling is done when the output power required exceeds that available from a single device, otherwise it may be better to use a higher rated device, depending on cost, availability, reliability etc.
For bipolar devices the input impedance is typically just a few Ohms, eg. the Motorola MRF858S S11 at 900 MHz is 0.94 angle165° giving Zin = 1.5 + j7 Ohm, but at rated output power (3.6 W) it is 1.2 + j3.5 Ohm. Connecting several in parallel would therefore create even lower impedances so it is normal to transform each to an intermediate impedance of say 10 Ω to 25 Ω before combining.
Devices may be directly connected in parallel but this approach has several disadvantages:
Parallel Combining of MOSFETs
MOSFETs can be paralleled but their gates require to be isolated from each other otherwise there is a risk of oscillation caused by the internal capacitances resonating with the gate source bonding wires. The effective Q may be reduced by adding a resistor in series with each gate connection. However this will reduce the high frequency performance typically to around VHF because of the RC or LC filter formed by this resistor and the internal strays.
Advantages: High power can be generated by 2 or more devices, no intermediated impedance matching is required and good power sharing can be obtained.
Disadvantages: High frequency performance limited by adding necessary gate isolation resistors.
Combining 2 or More Amplifiers
A common way of combining amplifiers is to use a suitable splitter at the inputs and a reciprocal combiner at the outputs. Examples of splitter/combiners are the Wilkinson, Lange coupler and branch line coupler (BLC). The Wilkinson can be theoretically designed with more than 2 ports but manufacturing can become challenging. The 2-port Wilkinson, Lange or BLC can be cascaded to provide 2N ports. The Lange of reasonable dimensions is only suitable for microstrip circuits at microwave frequencies. The Wilkinson and BLC can also be designed using lumped components for frequencies up to a few hundred MHz for good designs.
The splitters/combiners have isolated loads which absorb power from reflections at the input and output of each device thus masking them from the main input or output for the pair. The addition of compensated broadband matching for example is easy though some of the gain is forfeited. However the input signals will combine in phase making the combined IP3 and P1dB of two identical amplifiers 3 dB greater than the corresponding parameters of one of them. (The figure of 3 dB assumes that the output combiner loss is exactly theoretical (again 3 dB). Any actual loss more than theoretical would decrease the 3 dB improvement accordingly. For example if the actual combiner loss was 3.5 dB the IP3 and P1dB improvement obtainable would be 2.5 dB.
Two or more amplifiers may be combined in differing phases, for example in quadrature (a 90° phase relationship) or in anti-phase (a 180° phase relationship). In both cases, the phase shift of the splitter must be corrected for (reversed) in the combiner. These give advantages in terms of distortion products by cancelling even order products and odd order products respectively. In practice the products are not literally cancelled but attenuated appreciably.
Although it is of less concern usually at high power, for two identical devices, a loss free splitter and a loss free combiner, the overall noise figure of the combined network will be identical to the noise figure of one of the devices. The overall gain would also be identical to the gain of one of the devices. Once losses are introduced the overall noise figure would increase and the overall gain would decrease.