Audio amplifier current mirrors AC behavior

This investigation shows the effect of current mirror degeneration resistors on the AC behavior of different current mirrors. Further improvement of the AC response is discussed as well.

The background of this investigation is my observation that different degeneration of a current mirror being part of an audio amplifier may significantly deteriorate stability of an otherwise perfectly stable amplifier.

This investigation was preformed in simulation only using LTspice.

The input current is 10mA for all mirrors.

The intial investigation used only the BC546B transistor model. Due to observed different behavior of different transistor models, further transistor models were added: BC337-40 and BC547C.

The degeneration of the current mirrors was stepped exponentially from 25Ω to 3200Ω, i.e. 25Ω, 50Ω, 100Ω, 200Ω, 400Ω, 800Ω, 1600Ω and 3200Ω. The higher resistor values are impractical, but illustrate the trend beyond the typical range.

The illustration shows the AC response with quiescent current increased by 10kΩ and 2kΩ. The quiescent current is a function of the ratio between resistor R_d and the resistor R_q. For illustration I just stepped R_d with two fixed values of R_q. It appears as if the exact value of R_q does not have big impact on the AC response. Any increase of the EF transistors I_q is an improvement.

With BC546B transistors, the AC response shows lower peaking of the amplitude with increased quiescent current through the emitter follower transistor. Also, the bandwidth of the current mirror is increased. Ideal values of R_d would be 100Ω, which yields improved bandwidth over the standard EF current mirror with 200Ω degeneration.

With the BC337-40 models, the mirror shows slightly improved bandwidth, but also more pronounced gain peaking, that can be tamed using rather high emitter degeneration resistors, which cuts into bandwidth in turn.

Given that enough voltage headroom can be sacrificed, higher degeneration of current mirrors seems desirable because transistor mismatch becomes swamped and noise is lowered with higher degeneration. Higher degeneration in turn adds a significant pole to the current mirror transfer characteristic, which may impact overall amplifier stability. Some methods to mitigate this pole have been investigated and may be useful to mitigate the pole while maintaining the benefits of high degeneration. Too low degeneration in turn results in amplitude peaking in some kinds of current mirrors with some transistors.

Different mirrors may behave differently dependent on the transistor used. This means that not only the mirror topology is relevant. The best example is the four and five transistor Wilson mirrors, that show very different AC response dependent on the transistor model.

In a practical implementation there are further factors that influence the AC behavior of any current mirror like the output impedance of the current source feeding into the current mirror and also the load attached to the current mirror. Any capacitive load significantly alters the AC response. I encountered this effect when designing an audio amplifier: Addition of a R-C-snubber at the output of the emitter follower current mirror resulted in a gain peak whose amplitude grew with increased load capacitance. The five transistor Wilson mirror seemed to best suit the given application in my case despite showing a rather ill AC response on its own.

Apart from real issues with the EF current mirror mentioned earlier, I encountered instability with the Wilson mirror in simulation and this is why I expanded the investigation to include AC response for mirors using the BC337-40 transistor model.

When I simulated a CFA using a four transistor Wilson mirror, I observed severe open loop and closed loop gain peaking after changing the transistor model. The amplifier simulated fine with BC547C models for the mirror, and once I changed the mirror transistor models to BC337-40, a notable gain peak appeared beyond the amplifiers normal bandwidth. This even manifested in the transient response simulation as sustained low amplitude oscillation in the tens of MHz range atop the amplified signal.

Such local gain peaks of circuit blocks cause instability that cannot be remedied compensating the global feedback loop of an amplifier. It is therefore advisable to investigate and optimize each circuit block on its own to ensure proper operation. Once inside a complex assembly like an amplifier, it may become difficult to correctly identify the root cause of instability. Excessive compensation of the global feedback loop may be applied, but this is futile since the real root cause is different.