15

FETs versus BJTs: the linearity competition

This short chapter was published when debate about the linearity of power FETs was raging, or at any rate smouldering, in the letters columns of Electronics World. Many contributors were content to point out that FETs must be more linear than bipolar transistors because everyone says so. This sort of argument has never had much appeal for me, and I carried out several investigations to see if there was any way in which FETs could be claimed to produce less distortion. One of the problems with this is making a meaningful comparison between two rather different kinds of active device. Here I try to level the playing field by making the transconductances the same; the BJT wins heavily on linearity when degenerated to have the same low transconductance as a FET. In a complete power amplifier, the situation is naturally rather more complex; BJT output devices need BJT drivers (I suppose you could use FET drivers, but I think it a most unpromising route to head down) which introduce more distortion than you might expect, while many FET power output stages can dispense with drivers altogether, so long as the VAS is capable of charging and discharging those rather large gate-capacitances. Nevertheless, I am confident that in a fair contest a BJT amplifier will always have lower distortion than its FET equivalent. In particular, it will have lower and less nasty crossover artifacts.

Not everyone felt that this contribution settled the matter for good – the SPICE models used to simulate the FETs were the focus of particular attention. Fortunately, since then a lot more has been published on FET models, and it appears my conclusions were correct.

There has been much debate recently as to whether power FETs or bipolar junction transistors (BJTs) are superior in power amplifier output stages. Reference 1 is a good example. It has often been asserted that power FETs are more linear than BJTs, usually in tones that suggest that only the truly benighted are unaware of this.

In audio electronics it is a good rule of thumb that if an apparent fact is repeated times without number, but also without any supporting data, it needs to be looked at very carefully indeed. I therefore present my own view of the situation here, in the hope that the resulting heat may generate some light.

I suggest that it is now well-established that power FETs, when used in conventional Class-B output stages, are a good deal less linear than BJTs.2 Gain deviations around the crossover region are far more severe for FETs than the relatively modest wobbles of correctly biased BJTs, and the shape of the FET gain-plot is inherently jagged, due to the way in which two square-law devices overlap.

The incremental gain range of a simple FET output stage is 0.84 to 0.79, range 0.05, and this is actually much greater than for the bipolar stages in Ref. 2; the emitter-follower stage gives 0.965 to 0.972 into 8 Ω, with a range of 0.007, and the complementary feedback pair gives 0.967 to 0.970 with a range of 0.003. The smaller ranges of gain-variation are reflected in the much lower THD figures when PSpice data is subjected to Fourier analysis.

However, the most important difference may be that the bipolar gain variations are gentle wobbles, while all FET plots seem to have abrupt changes. These are much harder to linearise with negative feedback that must decline with rising frequency. The basically exponential Ic/Vbe characteristics of two BJTs approach much more closely the ideal of conjugate mathematical functions – i.e. always adding up to 1. This is the root cause of the much lower crossover distortion.

Close-up examination of the way in which the two types of device begin conducting as their input voltages increase shows that FETs move abruptly into the square-law part of their characteristic, while the exponential behaviour of bipolar devices actually gives a much slower and smoother start to conduction (see Figures 4 and 5).

Similarly, recent work shows that less conventional approaches, such as the common-collector/common-emitter configuration of Bengt Olsson, also suffer from the non-conjugate nature of FETs. They also show sharp changes in gain. Gevel3 shows that this holds for both versions of the stage proposed by Olsson, using both N and P-channel drivers. There are always sharp gain-changes.

Class A stage

It occurred to me that the idea that FETs are more linear was based not on Class-B power-amplifier applications, but on the behaviour of a single device in Class-A. You might argue that the roughly square-law nature of a fet’s Id/Vgs law is intuitively more ‘linear’ than the exponential Ic/Vbe law of a BJT, but it is difficult to know quite how to define ‘linear’ in this context. Certainly a square-law device will generate predominantly low-order harmonics, but this says nothing about the relative amounts produced.

In truth the BJT/FET contest is a comparison between apples and aardvarks, the main problem being that the raw transconductance (gm) of a BJT is far higher than for any power FET. Figure 1 illustrates the conceptual test circuit; both a TO3 BJT MJ802 and an IRF240 power FET have an increasing d.c. voltage, Vin, applied to their base/gate, and the resulting collector and drain currents from PSpice simulation are plotted in Figure 2.

f15-01-9780750681667
Figure 1 Linearity test circuit. Voltage Voffset adds 3V to the d.c. level applied to the FET gate, purely to keep the current curves helpfully adjacent on a graph.
f15-02-9780750681667
Figure 2 Graph of Ic and Id for the BJT and the FET. Curve A shows Ic for the BJT alone, while Curve B is the result for Re = 100 m Ω. The curved line is the Id result for a power FET without any degeneration.

Voltage Voffset is used to increase the voltage applied to FET M1 by 3.0 V because nothing much happens below a Vgs of 4 V, and it is helpful to have the curves on roughly the same axis. Curve A, for the bjt, goes almost vertically skywards, as a result of its far higher gm. To make the comparison meaningful, a small amount of local negative feedback is added to Q1 by Re. As this emitter degeneration is increased from 0.01 to 0.1 Ω, the Ic curves become closer in slope to the Id curve.

Because of the curved nature of the FET Id plot, it is not possible to pick an Re value that allows very close gm equivalence; a value of 0.1 Ω was chosen for Re, this being a reasonable approximation; see Curve B. However, the important point is that I think no-one could argue that the FET Id characteristic is more linear than Curve B.

This is made clearer by Figure 3, which directly plots transconductance against input voltage. There is no question that FET transconductance increases in a beautifully linear manner-but this ‘linearity’ is what results in a square-law Id increase. The near-constant gm lines for the BJT are a much more promising basis for the design of a linear amplifier.

f15-03-9780750681667
Figure 3 Graph of transconductance versus input voltage for BJT and FET. The near-horizontal lines are BJT gm for various RE values.
f15-04-9780750681667f15-05-9780750681667
Figure 4 and 5 Top are curves for a bipolar complementary feedback pair, crossover region ± 2 V, Vbias as a parameter. Fourth curve up provides good optimal setting – compare with curves below, for a FET source follower crossover region with ± 15 V range.

To forestall any objections that this comparison is nonsense because a BJT is a current-operated device, I add here a small reminder that this is untrue. The BJT is a voltage operated device, and the base current that flows is merely an inconvenient side-effect of the collector current induced by said base voltage. This is why beta varies more than most BJT parameters; the base current is an unavoidable error rather than the basis of transistor operation.

The PSpice simulation shown was checked against manufacturers’ curves for the devices, and the agreement was very good – almost unnervingly so. It therefore seems reasonable to rely on simulator output for these kind of studies; it is certainly infinitely quicker than doing the real measurements. In addition, the comprehensive power-FET component libraries that are part of PSpice allow the testing to be generalised over a huge number of component types without you needing to buy them.

To conclude, I think it is probably irrelevant to simply compare a naked BJT with a naked FET. Perhaps the vital point is that a bipolar device has much more raw transconductance gain to begin with, and this can be handily converted into better linearity by local feedback, i.e. adding a little emitter degeneration.

If the transconductance is thus brought down roughly to FET levels, the bipolar has far superior large-signal linearity. I must admit to a sneaking feeling that if practical power BJTs had come along after FETs, they would have been seized upon with glee as a major step forward in power amplification.

References

1. Hawtin V. Letters. In: December 1994:1037. EWWW.

2. Self D. Distortion in power amplifiers’. EW + WW. November 1993;932–934 Part 4.

3. Gevel M. Private Communication. January 1995.

..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset
3.81.184.170