GB2478585A - An RF amplifier linearised by RF feedback, and having a loop filter resonator of enhanced Q - Google Patents

Abstract

An RF amplifier 2 is linearised by RF feedback 4,6. The centre frequency Q of a resonator in the loop filter is adjusted by a control loop 10, 11,26 so that the resonator centre frequency follows the frequency of the input RF signal 1. The Q of the resonator, which is enhanced by local positive feedback, is similarly optimised by a control loop 10, 11,27. The low Q factor of integrated circuit resonators can be enhanced. The arrangement is much simpler than a Cartesian feedback circuit. High linearity and stability is achieved at low cost.

Description

"Improved Band-Pass Feedback Amplifier Circuits" This invention relates to a method of improving the performance of band-pass feedback amplifier circuits.

As is well known, Class A amplification can achieve good linearity at the cost of low power efficiency. However, as all modem modulation techniques involve the use of non-constant-envelope signals, and indeed multi-carrier systems, whether code division multiplexed, frequency division multiplexed or OFDM, all of which have a high peak-to-mean power ratio and require high linearity. This conflicts with the demands for higher transmifter power efficiency brought on by rising energy costs and the requirements to reduce C02 emissions. Techniques exist to increase the ratio of peak to mean power which can be efficiently amplified, from Class B techniques through to the Doherty technique, CALLUM and the variation of supply voltage in anticipation of signal peaks. However these all themselves generate signal distortion. Coping with these demands requires special linearisation techniques, such as pre-distortion techniques, feedback techniques and feed forward techniques. In the case of pre-distortion techniques, the implementation can be complex, and therefore costly, because of the need to monitor the transmission over a large bandwidth in order to adjust the pre-distortion. Also these techniques can only be used to compensate for transfer characteristic non-linearity, and do not reduce mutual intermodulation effects. between co-located transmifters due to coupling between transmitter antennae. Feed forward techniques can be used effectively, although they require accurate adjustment making them expensive to produce, as well as not reducing mutual intermodulation effects.

Another limitation is that they require the use of two amplifiers, one amplifying the signal, the other amplifying the error in the output of the main amplifier. A 3dB compression of the main amplifier would require an auxiliary amplifier at least as large as the main one. And if the outputs are not to affect each others', their output signals must be combined together without mutual interaction which requires a hybrid combiner which has an inherent loss of 3dB if main and error signal are uncorrelated.

Feedback is a well established technique for improving amplifier consistency and linearity at lower frequencies and was the first technique to be tried for the linearisation of broadband rf amplifiers. However it is difficult to achieve stability of amplifiers with a large open loop gain over large bandwidths. However, with transmitter amplifiers, although the bandwidth of the signals to be amplified may be large, it is normally still small compared to its resonant frequency. Band-pass feedback may be used as effectively in this situation as in the usual situation where the signal bandwidth is at baseband.

In order to achieve this band-pass loop filtering is used. For each pole or zero of a baseband filter, a complex conjugate pair is used for an equivalent band-pass filter, with the same real part but with imaginary parts displaced up and down by the amount of the resonant frequency. For band-pass filters of low fractional percentage bandwidth, the complex conjugate sets of poles and zeros are sufficiently separated for the frequency response at frequencies reasonably near to the pass-band, to a close approximation, to be frequency translated versions of the response of the baseband filter, apart from a scale factor. This scale factor will in general include a phase offset at the resonant frequency. This is caused by the the presence poles or zero's on the real axis, inevitable propagation delay, the effect of the conjugate set of poles and parasitics. In a feedback loop, it is important that any phase offset at the resonant frequency is compensated for by addition of an appropriate phase shifting network. A small error in this will result in asymmetry of the closed loop poles about the resonant frequency. This asymmetry may be seen by an assymmetry in the noise floor of the amplifier output. The worst effect of this is caused by the positions of the the real parts of the pole positions. When the phase error is sufficiently large one of the poles will cross the real axis and the loop will oscillate. Measurement of this asymmetry could be used for a closed loop control of the of loop phase compensation. However, if the loop were oscillating, the presence of a large oscillation limited only by non-linear effects, would make it difficult or impossible to make meaningful measurements. Adjustment could be made on the basis of whether the oscillation were above or below the filter resonant frequency, but by this time a self-tuning resonator, as in this invention would have moved from the correct value causing havoc! Tt is better to avoid the oscillation in the first place as is achieved using the open loop control described in Prior art Patent W00189081.

Linearising power amplifiers is unlikely to require a ioop filter more complex than the band-pass equivalent of dominant pole compensation such as is used with operational amplifiers. Rosen and Owens published a paper in which they describe such a linearized amplifier. (Power Amplifier Linearity Studies for SSB Transmissions, Rosen, H.; Owens, A., IEEE Transactions on Communications Systems Volume 12, Issue 2, June 1964 Page(s):150 -159) Although this worked fine in the lab, as the magnitude of the imaginary parts of pole locations is increased by the frequency translation while the relative positions and real parts of these locations are unchanged. This means that the values of Q (defined as twice the ratio of imaginary part to real part of the pole position, or equivalently the ratio of energy stored to energy dissipated per cycle in resonance) and the absolute positional accuracy of pole positions must be increased in proportion to the resonant frequency. This can pose a practical implementational challenge.

This was the main reason why the idea was replaced by a variant on it in which the feedback signal was translated to a complex baseband signal at which the use of an integrator provides the effect of a single pole pair resonator of infinite Q, this being the ideal. This technique of Cartesian feedback has be successfully used both with analogue and digital loop filtering. The advantages of this technique are the ease of implementing the filters and the ease of enabling the transmitter to operate over a band of frequencies without requiring manual tuning of high Q cavity resonators and adjustment of both the resonant frequency and phase offset. The former could be implemented by changing the frequency of the conversion oscillator, and the latter by changing the gain coefficients of cross coupling terms between the real and imaginary signals. The technique has been used very successfully with narrow band modulation i.e. a few kHz, but is unsuited to the wider bandwidths of a few MHz now employed in order to achieve high data rates with OFDM and CDMA.

Analysis shows that there is an ultimate limit to the loop gain bandwidth product of feedback amplifiers, which limits the bandwidth over which a specified amount of linearisation can be achieved. This is determined by the average group delay round the loop excluding the effect of the dominant pole over the frequency range over which the magnitude of loop gain exceeds unity.

Some improvement is possible using second order complex filtering, but with greater complexity and fragility, and would be more effective in increasing linearisation to normally unnecessary levels than increasing the signal bandwidth with the level of linearisation normally required. But the use of any more poles gives no significant benefit.

The extra cost, and, more importantly, the reduction in linearisation bandwidth due to the that the extra delays in the Cartesian ioop, particularly that of the two frequency conversion processes, especially the filtering required and the group delays of baseband signal processing, especially if performed digitally,. led to the original idea being revisited and an improvement described in UK patent 1312159 and W00189081.

This is a radio frequency feedback amplifier circuit of high linearity within, and close to, the range of frequencies with which the circuit is to be used, the circuit comprising high gain amplifier circuit incorporating a band-pass filter in the form of one or more resonators connected in the forward path of the amplifier means and having its resonant frequency at substantially the signal frequency, and feedback means in the form of a linear passive circuit.

Such an amplifier circuit may be used to achieve high linearity and stability at reasonable manufacturing cost, with much greater simplicity than can be achieved using other linearisation techniques.

As used in this specification the term "resonator" includes any band-pass filter arrangement having a single dominant conjugate pole pair. II is worth noting that the input and output of a resonator can be signals propagating in different directions at the same physical point. These may comprise passive components, such as inductors and capacitors, cavities or ceramic resonators or MEMS devices of various kinds, YTG resonators or may rely on active components as well. An example of the latter are state variable resonators which comprise two integrators in a feedback ioop. The integrators can be implemented using operational amplifiers with capacitive feedback, but at r.f more commonly a transistors acts as a transconductance amplifiers charging a capacitance known as a gm-C integrator. It is clear then that the resonators can be used to implement the band-pass equivalent of each pole in a base-band filter.

It should be noted that the technique of band-pass feedback is qualitatively different to simple rf feedback such as is commonly used for impedance matching and gain stabilisation in MMXC's. This latter requires instantaneous waveform following involving a wide range of spectral components. Such techniques as class B amplification cannot be used as they are likely to result in waveform distortion due to such effects as slew rate limiting and saturation, which cannot be ameliorated by feedback. By contrast band-pass feedback, like Cartesian feedback, is only concerned with the instantaneous complex envelope of the signal. This removes the limitation of using class A amplification.

In order to achieve the greatest accuracy at baseband, highest loop gain should be sought within the signal bandwidth. Baseband servomechanisms generally use of one or more integrators (whose gains go, theoretically, to infinity at zero frequency) in the forward path, to maximise the ioop gain over the signal bandwidth. The maximum number of such integrators which can normally be used while ensuring stability is normally three, though over the bandwidths used in the applications of this invention, the effect of parasitics make it difficult to achieve stability with more than two, and often good performance can be achieved with only one. The replacement of integrators by first order lag circuits (whose gain rises to some limiting value as frequencies tend to zero) limits the ioop gain at the lowest frequencies, but has little effect on the stability or performance of the loop above the break-point frequency of the lag.

Translating to band-pass feedback, we replace the integrators by resonators at the resonant frequency, ideally having infinite Q, i.e. with poles lying on the imaginary axis. The requirement that the break-point frequency is much less than the signal bandwidth translates to the requirement that the 3dB bandwidth of the resonators should be much less than the signal bandwidth.

What is apparent using techniques such as this, is that both the frequency and the Q of the resonator should be accurately set, the former to the centre of the signal pass band and the latter to the largest value possible. It should be noted that a resonator of 2 MHz 3dB bandwidth at 2GHz would have a Q of 1000. This 2MHz a fairly broad band signal and very much higher values of Q will be desirable for many applications.

Cavity resonators can achieve these values of Q, but such resonators are large, expensive and difficult to tune over a significant bandwidth, and are frequently accompanied by zeros close to the resonant frequency pole which can compromise the filtering performance, though not necessarily the stability in a feedback loop.

All practical resonators, though, have finite value of Q. In the case of passive resonators this is due to the losses in the resonator itself and to the loss due to the coupling to the input and output ports. The ratio of these determines the insertion loss of the filter. If the loaded Q is too high the insertion loss will be high and this reduces the ratio of signal passing through the filter to that which may by-pass it due to passive parasitic coupling. This latter degrades the ultimate stop band rejection of the filter, which in turn reduces the ioop gain achievable. A familiar example of this is the admittance of quartz crystals. The capacitance by-passing the resonator is responsible for a parasitic admittance zero close to the admittance zero of the crystal itself As well as this, the insertion loss must be compensated for by additional gain stages which add to the ioop delay, which in turn degrades the performance of the feedback ioop as will be seen below.

In the case of state variable resonators, reduction in Q results from the series resistance of capacitors. At low frequencies this is usually very small compared to the reactance of the capacitors, resulting in very high values of Q. However the loss resistance of capacitors at UHF and above is very significant. Typical values for standard surface mount components are usually less than 100, and on-chip capacitors an order of magnitude less. Thus the assumptions frequently made in analysis of negligible loss, are not valid at these frequencies, especially in band-pass filters where the required values of Q are scaled up, as described above, inversely as the fractional bandwidth., These important issues are totally neglected in much of the work on bandpass filtering, for example eq 2 in US patent 5729230 by Jensen, Gopal and Cosand.

The use of positive feedback, or regeneration, provides a means of increasing the Q of resonators. In the case of resonators consisting of two integrators, such as the gm-C circuit, positive feedback round one of these is one means of achieving this effect.

Parasitic delays in this type of circuit may inadvertently have such a regenerative effect, which may help in this application.

Where the resonator is a passive tuned circuit a transistor, tunnel diode or other amplifying device can provide the positive feedback using any of a number of circuit topologies corresponding to familiar oscillator circuits. Careful design is, of course, required to achieve appropriate levels of signal handling and independent control of frequency and regeneration.

However in both types of regenerative resonator, the real part of the pole position is determined by the balance of damping and positive feedback. On its own, a regenerative resonator with a pole with a positive real part, i.e. on the right hand side of the imaginary axis, is unstable and will oscillate. However in a stable outer feedback loop the pole positions are moved to the left and stability can result so long as the loop gain is sufficiently high but not too high. This mathematical result is confirmed by the inventor has never experienced any stability problems due to poles on the right hand side whether with band-pass amplifiers or riding a bicycle! However the loop gain at mid-band is degraded by the same amount by poles at a particular distance from the imaginary axis whether to the left or the right. Again, this is the same for band-pass feedback loops as for baseband ones.

It is therefore desirable in order to implement such feedback amplifiers to employ some some method of controlling both the resonant frequency and the Q of the resonators to their optimal values, i.e. the resonant frequency to the centre of the passband and the Q to the largest achievable value. II is instructive to view this geometrically as the placement of the conjugate poles of the resonator in a complex plane.

Prior art Patent WOO 189081 A discloses means to control the imaginary part of the pole position of the loop gain of a band-pass feedback amplifier by measuring the phase of the resonator response by taking the product of the input and output of the resonator. However this does nothing to control the real part of the pole position which defines the Q of the regenerative resonator. The optimal value of this real part is zero, when the resonator is on the border between stability and instability, as this corresponds to the Q, and hence the loop gain going to infinity at the resonant frequency. This is of importance particularly for on-chip implementations where Q values are low, typically around 10, for any form of tunable on chip-resonator. It is worth noting though that even when a regenerative resonator has poles in the right hand half plane, this need not affect the stability of the overall feedback loop so long as the loop gain is above a certain minimum value. This is illustrated in the root locus plot figure 5. The effect of the resonator having poles in the right hand half plane is that the ioop gain is reduced causing an increased error vector, the result of which is, perhaps unexpectedly, that the closed loop gain is higher than the correct value. This is because this error is in the same phase as the correct output, as opposed to being in anti-phase, as normally happens when the loop gain is too low, or at right angles, as when the resonant frequency is incorrect. This makes manual adjustment difficult as this cannot be done by adjusting to achieve maximum output, as works well when only adjusting resonant frequency. The output power just increases monotonically as regeneration is increased. In fact you cannot even do it by minimizing distortion products as these often depend critically on output level.

This patent teaches an electronic feedback circuit which can automatically adjust to their optimal values both the resonant frequency fo-and the Q of a resonator such as is required for band-pass feedback loops such as the amplifiers described above.

In order that the invention may be more fully understood, preferred embodiments in accordance with the invention will now be described, by way of example, with reference to the accompanying drawings in which: Figure 1 shows a block diagram of a prior art circuit; Figure 2 shows a root locus plot of a type 1 feedback circuit; Figure 3 shows a root locus plot of the band pass equivalent of the circuit whose root locus is shown in figure 2; Figure 4 shows the block diagram of a preferred embodiment of a regenerative resonator Figure 5 shows the root locus of a bandpass feedback amplifier using a regenerative resonator with a pole in the right hand half plane; Figure 6 shows a block diagram of an embodiment of the invention; Figure 7 shows a pole zero diagram showing the operation of the invention; Figure 8 shows a block diagram of an embodiment of the invention 1. Figure 1 shows a block diagram of a radio frequency (rf) feedback power amplifier circuit according to Prior art Patent WOO 189081 A in which the input signal, 1, is supplied via a comparator, which could be a hybrid signal combiner 6 used as a comparator, to a resonator 7 incorporating a means of providing tuning of the resonant frequency fo typically by means of varactor diodes. A phase detector 8 is provided to measure the phase difference between input, 14, and output, 13, of the resonator 7, and provide an error signal which is applied to an integrator 7 which supplies a control signal, 27, to the resonator to adjust the resonant frequency to provide a phase shift at the centre of the rapid 180 phase swing which occurs at resonance, between the inputs of the phase detector, 8, at the transmission frequency. The phase shifts and delays in the circuit are adjusted or compensated in the design so the the integrator input is zero at the resonator resonant frequency to cause the maximisation of the resonator gain at the signal frequency. It should be noted that this is not very critical, though, as even an error of 45° will only result in a reduction of 3dB in the loop gain.

The output of the resonator 7 is fed to the rf power amplifier stage 2 which in turn supplies its output, 3, to the transmit aerial. A, preferably directional, coupler 4 is provided so as to supply a feedback signal to the comparator 6 by way of an electrically controllable attenuator and phase adjusting circuit 5 which is used to ensure that the phase of the loop gain is of the correct magnitude and close to 180 degrees at the resonant frequency. This is normally best implemented as an open loop control with the phase advance derived by function implementing circuit 12, from the resonant frequency control signal described above. Although this loop gain and phase adjustment circuits are shown in Prior art Patent wool 89081 A a being in the feedback circuit, there are some slight advantages in locating these, as is normally done, in the forward part of the loop in stead as in Figure 6.

Such a circuit provides high linearity using band-pass feedback, but without demodulation or frequency changing as would be required for Cartesian feedback. Since the performance of feedback loops at radio frequency is determined by the linearity and group delay round the loop, the removal of non-linearity and group delay associated with the components used in frequency changing etc., particularly any filtering, ensures that the methods described herein can achieve higher bandwidth and loop gains.

In such a feedback amplifier having a high Q resonator in the forward path of the high ioop gain feedback amplifier which would otherwise be broadband, closed loop stability is achieved so long as other phase shifts around the ioop contribute less than 90° within the frequency range over which ioop gain exceeds unity. If the circuit is adjusted so that the extra phase shift is zero at the centre of the frequency band, all that is required is that the phase shift of the circuit apart from the resonator should vary by less than 90° over a frequency range of the required ioop gain multiplied by the bandwidth over which the loop gain exceeds this value.

For example, if the extra phase shifts vary by less than 90° over say 50 MHz, then a ioop gain exceeding 20 dB is achievable over a bandwidth of 5 MHz or a loop gain of 40 dB over 500 kHz, etc. There is thus a loop gain bandwidth product of 50 MHz. This may also be specified as the average group delay of the open ioop, apart from the resonator, over the closed loop bandwidth which is 1/4*50MHz = 5 ns. This highlights the pivotal importance of keeping the group delay round the loop low. The linearity improvement at a particular frequency for small amounts of distortion is, as is well known, is given by the magnitude of the loop gain plus one. For large distortions though, particularly when well into saturation, the effective loop gain is reduced, which reduces the linearisation to some degree..

Figure 4 shows a preferred embodiment of a regenerative resonator wherein the resonant frequency of a passive resonator 20 is controlled by a control signal, 26. The output of this resonator is amplified by an amplifier 21, whose output is the output of the regenerative resonator, 25. This signal is fed back through an attenuator, 22, whose attenuation is controlled by another control signal, 27, to a summing circuit, 23. This summing circuit adds the output of attenuator, 22 to the input signal, 24. In so doing the Q of the resulting resonator is increased by an amount controlled by signal 24. If the phase of the output of the attenuator has been adjusted to be the same as that of the input signal, 24, this adjustment has no effect on the resonant frequencyfo but any error will cause some effect, and as there will always be some variation in Q as fo is adjusted, there will always be some small interaction between these two controls.

Figure 6 shows an embodiment of the invention illustrating the principle of the invention. This is that, whereas in the prior art illustrated in Fig 1, a single measurement on the transfer function of the resonator, with aforementioned phase adjustment or compensation, the phase, or equivalently its real part, provides the error signal used to adjust fo, the invention teaches the use of two measurements, phase and amplitude, or more conveniently real and imaginary parts of the said transfer function to act as error signals to adjust both fo and Q, or equivalently, real and imaginary parts of the conjugate pole pair positions. It is evident that if the pole lies exactly on the imaginary axis at the excitation frequency the resonator will have infinite gain, or equivalently, and more usefully, the reciprocal of the resonator gain will have zero values for both real and imaginary parts. The benign nature of such a control ioop, counter-intuitive though it may be, will become apparent in the following analysis of the embodiment illustrated in figure 6. This shows a bandpass feedback amplifier comprising a differencing circuit, 6, which provides the input to a regenerative resonator 7, whose output is scaled and phase rotated by circuit 5, as described for Figure 1, and then provides input to an amplifier, 2, a proportion of whose output is sampled by means of coupler 4, which provides the feedback signal which is subtracted from the system input signal, 1, by comparator 6.

If the feedback loop is stable, and the loop gain is much more than unity, which will be the case at frequencies near to j. the amplifier output signal will be almost independent of variations of fo. Tt is the error signal, 14, which will vary with the tuning of the resonator. It is therefore logical to take 13 as the reference signal and 14 as the signal input to a quadrature detector, 10, rather than vice versa.

There is also merit in using a limiting amplifier stage, not shown separately in Figure 6, to make the conversion gain of the detector, 10, independent of the signal level if this is not adequately provided for within the circuit of 10 already. Effective limiting will be assumed in the subsequent analysis as without it performance is degraded especially at low signal levels.

In order to understand the operation of the invention more clearly, let us perform a simple mathematical analysis of the circuit illustrated in figure 6 assuming the input signal, 1 is sinusoidal, the error signal 14 is represented by R ( Ve) , the reference signal 13 by t (V,) , the transfer function of the resonator, 7, by Ks Ks Hs s-p)(s-p) S +-S+CO..

where pjo and PJCo are the pole positions, or roots of the denominator.

V1 V1(jw-p)./co-p) ( . V. VjCo-p)I2K if jw-pco Hjco jwK The operation of the quadrature detector, 10, is such that it produces two outputs proportional to 9t(VeVr) and (er) respectively.

These outputs of the quadrature detector 10 are applied to the inputs of circuit 11 which performs the function of a square linear matrix. This can perform the functions of compensating for phase differences referred to above as well as providing linear compensation for the fact that adjusting fo will affect Q and adjustment of Q may affect resonant frequency fo., and compensating for the inevitable scaling differences between the adjustments of real and imaginary parts of p. Tf this is done, the adjustment is described by: =cV1(jw-p) This is illustrated in figure 7 in which 32 is the position of p, 31 is the position of jW, and the arrow 30 shows the direction of adjustment. Note that p is complex but c, the product of constants relating to the gains of 7,10,11 and 9 is ideally a real constant.

Ideally the coefficients of matrix circuit 10 would be adjusted to make the value of c real, which would mean that the control of fo and Q or real and imaginary parts of p would be independent first order loops with the same time constant. If not then the trajectory of p would spiral towards the same end point 31.

The fact that this is a simple linear relation means, amongst other things, that if the excitation is not sinusoidal, the adaptation will tend to converge to the centroid of the spectrum of whatever excitation is in fact used. This is particularly significant and useful for band-pass excitations such as modulated carriers.

In the event of that different adjustment loop gains, c, apply to the fo and Q loops, the system is not so elegantly described by complex numbers, and a full matrix formulation must be used. However the performance is not necessarily be compromised. Tndeed there may be benefits in so doing. In particular this may be useful, as may deliberate inserion of non-linearities, in modifying the trajectories to avoid p going too far to the right when making large step frequency changes which may invalidate the assumptions behind the linear system analysis above and result in instability or oscillation of the main feedback loop.

It is worth restating that the invention relates to a resonator within a stable feedback loop, and there is no reason that it could not be used for a multiplicity of resonators so long as they were all within a stable feedback loop. Without that feedback loop, though, the system would become unstable as soon asp crossed the imaginary axis.

One problem encountered in highly linear amplifiers such as described herein, is the fact that saturation occurs very rapidly under overload causing an abrupt increase in distortion. This would not matter too much were it not that efficiency generally decreases significantly as power levels are reduced below the maximum. The need to avoid overload in the worst case could cause a significant decrease in efficiency. There is, though, a simple solution to this shown in figure 8. This uses the fact that as a feedback loop overloads and is no longer able to track the input signal, 1, the error signal, which varies inversely with one plus the loop gain, rises dramatically as the effective loop gain falls. This can be used in a feedback loop to control the input level by means of an attenuator 42, controlled by a control circuit, 41 which could be an integrator, though there may be benefits in the use of a non-linear loop with faster attenuation rise-time than fall-time in many situations, whose input is derived from a signal level measuring device. There is merit in placing this at the output of the resonator rather than its input, as its spectrum is the same as that of the system output, apart from distortion, whereas resonator input signal will be multiplied by the inverse of the filter response the The controlled attenuator 41 is probably best performed using a PIN diode circuit, as these have good linearity, which is essential as it acts outside the signal feedback loop Another benefit of controlling the loop input by means of such an error monitoring signal in this way, would be to prevent the reduction in loop gain due to power amplifier overload from potentially causing loop instability if the poles of a regenerative resonator were too far into the right hand side to be compensated for by the reduced feedback loop gain as is mentioned in the discussion on figure 5 above.

Claims (5)

1. CLAIMS: 1. A radio frequency feedback amplifier circuit of high linearity at, and close to, the frequency of the input signal with which the circuit is to be used, the circuit comprising high gain amplifier means incorporating at least one band-pass filter in the form of a resonator connected in the forward path of the amplifier the resonant frequency of which may be tuned to the input frequency, an electrically adjustable positive feedback circuit to enhance the its Q of said resonator and negative feedback means in the form of a linear passive circuit.
2. 2. A circuit according to claim 1, wherein tuning circuit is provided for automatically tuning the resonant frequency and the Q of the resonator to optimal values in dependence on the input and output signals of said resonator..
3. 3. A circuit according to claim 1 or 2, wherein the tuning circuit comprises a quadrature detector whose signal and reference signals are proportional to the output and input signals of the resonator and a tuning circuit to tune both the resonant frequency and Q of said resonator.
4. 4. A circuit according to any previous claim, wherein a circuit is provided to adapt the loop gain of the feedback loop according to a measurement of system output noise.
5. 5. A circuit according to any previous claim, wherein an circuit is provided to adapt the loop gain of the feedback loop according to an open loop function of resonator centre.

   6 A circuit according to any previous claim, wherein a circuit is provided to correct the phase of the feedback loop according to a measurement of system output noise.

   7 A circuit according to any previous claim, wherein an a circuit is provided to adjust the phase of the feedback ioop according to an open loop function of resonator centre frequency 8 A circuit according to any previous claim, wherein a power control circuit is provided comprising a circuit to measure the signal level at the resonator output, an attenuator circuit at the system input and a control circuit control said attenuator circuit to increase attenuation should the output of said measuring circuit exceed some threshold.9. A radio frequency amplifier substantially as described above and illustrated in the accompanying drawings.