GB2438658A

GB2438658A - A phase compensation circuit for a low distortion audio amplifier

Info

Publication number: GB2438658A
Application number: GB0611016A
Authority: GB
Inventors: John Nigel Ellis
Original assignee: Individual
Current assignee: Individual
Priority date: 2006-06-03
Filing date: 2006-06-03
Publication date: 2007-12-05
Also published as: GB0611016D0

Abstract

An audio power amplifier has multiple stages to provide a high open-loop gain at high audio frequencies, which allows the reduction of gain error, gain doubling, crossover distortion and other amplifier distortions by application of feedback around the amplifier. Closed-loop stability is provided by a compensation network comprising a phase lead capacitor 23 in the feedback branch and a phase lag network 24,25 between the feedback and input nodes.

Description

Patent Application PLIL Method for Extremely Low Distortion in Audio

Amplifiers

Background

Audio amplifiers intended for faithful reproduction of recorded or transmitted sound material need to have very low levels of distortion to achieve near realistic performance levels. Many commercial amplifiers are content to offer distortion levels in the region of 0.1 to 0.4 %: indeed the early DIN standard set levels around this mark. New amplifiers using the Class D method in which devices are switched on and off rapidly tend to have distortion levels at around this mark despite theoretically being able to offer lower distortions as the switching method is supposedly not dependent on the linearity of the transistors used in the output stages.

It is recognised that for faithful reproduction of music it is necessary to minimise the effects of intermodulation distortion in particular. Intermodulation distortion is the phenomenon whereby any non-linearity in an amplifier can generate frequencies which do not exist in the input or signal source, when two or more input signal frequencies are present. For example, a 3 kHz signal and a 10 kHz signal can intermodulate to give 7 kHz and 13 kHz; and again further distortion products can arise from the intermodulation and harmonic distortion products. Thus, in the same example, an amplifier may generate a 20 kHz harmonic distortion signal from the 10 kHz input signal, which can intermodulate with 3 kHz to generate 17 kHz. All these components (7, 13, 17 kHz) are audible, and do not blend in with either of the original signals. Such distortion can therefore be quite detrimental to the listening experience. Since intermodulation products are generated whenever there is any non-linearity in an amplifier, to minimise these it is necessary to make an amplifier as linear as possible. If an amplifier were linear, no intermodulation nor harmonic distortion products can be generated.

Transistor amplifiers can be grouped into whether the output device or devices operate in Class A, in which they conduct during the whole of a cycle; Class B in which at least two stages are needed and in which each stage operates for half the input cycle; typically one responding to and providing the output for a positive going signal and the other responding to and providing the output for a negative going signal. Class B amplifiers are more efficient than Class A because they draw power only when an output is needed; but in practice Class B circuits are not linear in the region where one output stage crosses over into the other, leading to well-known crossover distortion. Most Class B amplifiers therefore are operated with a small bias in which they become Class AB. The low level Class A bias is intended to ensure that the two output halves operate smoothly: but in practice it is very difficult to remove the last traces of crossover distortion as the gains are never quite linear, even in small excursions, to provide a complete balance in the crossover region.

Other classes include Class C in which the output stage conducts for less than a full input cycle: it is unusual for an audio amplifier to operate in class C because the distortion levels generated tend to be very high, and are often controlled only through the use of a tuned circuit, in which a high Q factor can substantially restore a partial sinewave. Class D, in which the output stages are switched at a high frequency and modulated with the audio signals, should be able to provide low distortion, but in practice the switching transfer is not perfect either.

As a compromise between good power efficiency and low distortion, many amplifiers operate in Class AB mode. Thereafter it is necessary to minimise non-linear effects in order to minimise distortion. Bipolar transistors are frequently employed in input, intermediate and output stages to provide the amplification, but are inherently nonlinear as they respond with an exponential characteristic to the base-emitter voltage. Power MOSFETS can be used in the output stages of amplifiers today, but despite appearing to be more linear than a bipolar, actually are not because they suffer from transient effects more significantly. In any power transistor, heat dissipated in the device causes the temperature to rise. When the junction of a bipolar transistor heats up, its base-emitter voltage falls with a given current. This change can affect the slight forward bias appplied to reduce crossover distortion, and needs to be controlled. For MOSFETS, a temperature rise causes a gain reduction in the channel. MOSFETS normally exploit this to maximise the uniformity of current flowing within many small gates by causing a gain reduction which offsets the heating effect, and thereby achieves 100 % safe operating area Nevertheless, this gain reduction generates a reduction in the open loop gain, dynamically, which needs to be linearised.

Negative feedback is widely used to linearise an audio amplifier. Feedback can be applied locally or globally. In a series of papers, Otala described how transient distortion could be generated in a conventional amplifier4; how this could be minimised5; and a circuit to avoid the effect1. This prior art circuit exploited phase-lag at the input stage 1, and local phase-lead compensation 2 in the subsequent stages as shown in fig. 1.

In another prior art circuit2, a compensation capacitor is attached to the collector and base of a voltage amplifier stage node within the feedback loop of an amplifier as shown in outline in fig. 2. In such amplifiers, global feedback is applied from the output to the feedback node through a feedback resistor 3, of which a small proportion of the output voltage is actually generated by a voltage divider comprising 3 and 4. The stabilisation capacitor 5 can be seen to be internal to the feedback loop. The effect of such a capacitor, described originally by Miller, is to roll the open loop gain off. By so doing, the gain can be made to fall to unity at some high frequency at which the phase shift within the amplifier does not exceed the 90 degree shift caused by said capacitor. In this way an amplifier can be made unconditionally stable. The frequency response is set by the input stage gain (gm) and the said compensation capacitor.

In a typical bipolar transistor such as may be used in the output stage of an audio amplifier, the gain may vary by a factor of 2 or 3 over the operating current range. In turn, this means that the open loop gain must be sufficiently high that such changes in gain have minimal effect on the output signal. In practice, this means that the open loop gain must be high, so that the difference between, for example, an open loop gain of 2000 or 3000 and 1000 does not significantly affect the output. Normally the gain is determined from the standard feedback equation G= A 1+ AB where G is the gain with feedback, A is the open loop gain and B the feedback attenuation factor. If B were 0.05 then the closed loop gain would be 19.61 in an amplifier with an open loop gain of 1000 or 19.87 in an amplifier with an open loop gain of 3000. If a circuit exhibited this change dynamically, the distortion level would amount to about 1.3%. The said dynamic change in the gain of a MOSFET could be similar at 2:1. For this reason, open loop gains tend to be made much higher, and thus reduce non-linearities to lower levels. Modern transistors have been designed also to minimise the gain variation with current to minimise such non-linearities inherently.

Increasing the open loop gain has problems. Phase shifts within the amplifier mean that it is possible that unless some compensation were used, the phase shift could exceed 180 degrees while the gain remained higher than unity. In this case, the amplifier could oscillate. One of the main advantages of the Miller scheme is that the gain can be made to roll off monotonically throughout the audio frequency band up to and beyond the highest frequency required. For example, in a prior-art amplifier, it is possible to set the open loop gain to about 100,000. But phase shifts in the transistors through the amplifier stages mean that we may need to set the gain to unity at 10 MHz. Using a single roll-off element such as the said Miller capacitor means that the gain can only be 10 at 1 MHz and perhaps 30, a good compromise, Lfr at 300 kHz. Therefore this appears to be a reasonably good conceptual design, having a bandwidth of 300 kHz, audio frequency gain of 30 and unconditional stability. A simulation of such a model circuit reveals that at 20 kHz, there are crossover distortion products which cannot be controlled. The second harmonic distortion level is around 0.004%, 8, but the third harmonic is at 0.02%, 9, as shown in fig. 4. The distinguishing feature of such prior art circuits is that high order harmonics cannot be removed.

Another common method to compensate the frequency response in an amplifier is to connect the compensation capacitor from the pre-driver stage node as shown in fig. 3 but as a phase-lead configuration connected to the feedback node. This was used successfully in a simple amplifier6. The main problem with this approach, however, is that the gain is rolled down to unity but cannot be controlled below unity without some additional compensation. This is illustrated in fig. 5 which indicates that an input series resistor 7 is required in addition to the compensation capacitor 6 in order to prevent the spike 10 shown in fig. 5, measured across the input transistor emitter and base junction, which could give rise to oscillation.

In a further prior art design, the method shown in fig. 4 using phase lead compensation was combined with input-stage phase lag, forming the phase-lead, input-lag (PLIL) method as shown in fig. 6. The method used the input phase lag capacitor 11 to control the said high frequency spike 10 by rolling off the gain at high frequencies. However, because the phase lead capacitor 12 is taken from the collector of the internal voltage amplifier node 13 the crossover distortion harmonics arising in the subsequent stages (not shown) cannot be controlled.

3. The New Method In this new design, phase lead and input lag compensation can be applied to a simple design as shown in a typical embodiment in fig 6. A differential input pair of transistors 14 and 15 are used to set the output voltage rail 16 to zero volts under normal conditions. Feedback is applied through the feedback resistor 17 and attenuated with resistor 18 such that the desired gain of the amplifier is achieved.

The differential input pair feed a differential pre-driver stage comprising four transistors 19,20,21 and 22. Transistors 19 and 20 form a differential stage while 21 and 22 act as a current mirror. Transistors 20 and 22 form a push-pull stage which drives the driver and output transistors (not shown). The amplifier can be augmented with protection circuitry to prevent damage in the event that the output is short circuited.

Feedback from the output is stabilised with a phase lead capacitor 23 which is in parallel with the said feedback resistor 17. The amplifier would oscillate without the use of an input phase lag circuit made up from a resistor 24 and capacitor 25 in conjunction with a series input resistor 26. The effect can be seen in fig. 7 showing the frequency response and input differential emitter-base voltage with and without the input phase lag components (25,26,27,28).

In particular, the advantage for this configuration is that there are no phase-lag nor lead capacitors attached to the pre-driver stage at the collector of 20 or 22. This means that the pre-driver stage operates not as a voltage amplifier as it is commonly known by those practising the art but more of a true current amplifier. As a result, the

S

transition between the two output stage halves is smooth; distortion arising from the gain-doubling distortion, which arises in a conventional prior-art circuit when both output devices are on due to the said slight Class A bias, does not exist, and crossover distortion is virtually eliminated.

In this embodiment, the overall distortion levels at 20 kHz are typically 0.02% but these are not due to crossover effects but second and third harmonic effects arising from said high current non-linearities in the transistor. At I kHz and power levels of around 1W, typically the region where crossover distortion is around its maximum, the second harmonic distortion level is around 0.001%, 29, as simulated in fig. 8.

In a preferred embodiment, a three-stage amplifier is used. In fig. 9, a differential transistor pair 30 form the input and feedback connections, and hold the output rail at or near zero under quiescent conditions. The said differential pair is connected to a second differential pair 31. The said second differential pair are connected to a third differential pair 32. Said third differential pair are connected to current mirror pair 33.

Phase lag compensation is arranged at the input with capacitor-resistor network 34 and phase-lead compensation is applied around the feedback resistor in network 35.

Due to the extremely high gain it may be necessary to include additional phase compensation to stabilise the amplifier. For said purposes, capacitors 36 and 37 in the said current mirror loop; and small phase lag capacitors 38 and 39 are connected across driver transistors. The said driver transistors are connected to an output transistor stage and the common connection of said output transistors is connected to a loudspeaker 42.

The said input differential pair 30 may be fed from a constant current source 40 which may be made up from any conventional means including combinations of transistors and diodes. The said second stage differential transistor pair 31 are fed from a constant current source 41 which may be made up from any conventional means including combinations of transistors and diodes. The input connection to the said first stage differential pair may additionally include a series resistor 42.

It is evident to those practiced in the art that several alternative combinations of input and intermediate stage transistor circuits can be used. It is also evident that alternative means of stabilising against said high frequency oscillation may be achieved with additional components in series or parallel with the said phase lead and lag networks 36, 37, 38 and 39. It is also evident to those practiced in the art that other possible configurations may be used to optimise the frequency response in the said feedback network 34 with additional series or parallel components. It is also evident that further components may also be needed to provide a fine-tuning of said amplifier against said oscillation.

The said phase-lead, input lag circuit configuration develops distortion levels at 20 kHz and 50W as shown in fig. 10. The second, 43 and third harmonic, 44, distortion levels are 11 ppm; and higher harmonic distortion levels are below this. This is well below the distortion levels which prior art PLIL can achieve. At lower powers and frequencies, the distortion level is even lower. For example, at I W and 1 kHz the distortion is less than I ppm, 45, as simulated and shown in fig. 11. This demonstrates that in the critical crossover region, the distortion of the PLIL amplifier has negligible crossover distortion. (0

Because of the largely current-mode drive in the pre-driver stage, dynamic crossover distortion will also be very low provided that the output stage bias current ensures that the output stages are never both off at the same time. This requires a sufficiently large bias current to ensure this is the case at 20 kHz, the highest audio frequency which can be heard, but dynamic changes which increase or reduce this slightly have no material effect on the distortion. Therefore, special transistors with feedback diodes which may help conventional circuits are not required either.

References 1. "An Audio Amplifier for Ultimate Quality Requirements", Lohstroh and Otala, IEEE transactions AU-21 no.6, Dec. 1973, pp.545-557 2. Self, D. "Distortion in Audio Amplifiers"-?', Electronics World, Feb. 1994, pp. 137- 3. Ellis. J, "Audio Power Amplifier Frequency Compensation", Electronics World, March 2003, pp. 10-17 4. Otala, M, "Transient Distortion in Audio Power Amplifiers", IEEE Trans. AU-18, no3, Sept. 1970, pp. 234-239 5. Otala, M., "Circuit Design Modifications for Minimising Transient Intermodulation Distortion", Jnl,. Audio Engineering Society, vol. 20, no.5. June 1972, pp. 396-399 6. Bailey, A.R., "3OWHigt, Fidelity Amplifier", Wireless World, May 1968, pp. 94-98.

Report dated 29 May 2006 Original design concept dated 1st February 2006.