Chapter 5 Opamps for Low Voltages – Small Signal Audio Design, 3rd Edition

CHAPTER 5

Opamps for Low Voltages

High Fidelity From Low Voltages

Nowadays there is considerable interest in audio circuitry that can run from +5 V, typically obtained from a USB port. It is expected that reasonable quality will be obtained: what might be called five-volt fidelity. Since there is only a single +5 V rail available, the maximum possible peak voltage is clearly ±2.5 V, equivalent to 1.77 Vrms or +7.16 dBu. (This calculation obviously ignores opamp output saturation voltages.) A slew rate of only 0.31 V/μs is enough to give this maximum output at 20 kHz.

When the usual ±17 V rails are used, you get a maximum level of 12.02 Vrms or +23.81 dBu. Five-volt audio gives a maximum level that is 16.6 dB less, so the dynamic range is reduced by the same amount. If the same relative headroom is required, the nominal signal level will have to be correspondingly reduced by 16.6 dB, and the 5 V signal path will have a worse signal/noise ratio by 16.6 dB. Often a compromise between these two extremes is more appropriate. In this technology it is especially important to make sure that what dynamic range does exist is not compromised by deficiencies in design.

While +5 V was for many years the standard voltage for running digital circuitry, for some time now +3.3 V has been used to reduce power consumption and make larger processor ICs feasible. It is perhaps not very likely that applications will arise requiring quality audio circuitry to run off such a low supply, but the existence of opamps in Table 5.1 that are rated to operate down to 2.7 V and even 2.2 V shows that opamp manufacturers have this sort of thing in mind.

With a single +3.3 V rail, the maximum possible peak voltage is ±1.65 V, equivalent to 1.17 Vrms or +3.55 dBu, ignoring output saturation voltages. This is 3.61 dB less than for a +5 V rail, so the dynamic range issue is that much more acute. A slew rate of only 0.21 V/μs is enough to give the maximum output at 20 kHz.

The issues of +5 V operation are considered first, followed by those of +3.3 V operation.

Running Opamps From a Single +5 V Supply Rail

The vast majority of audio circuitry runs from dual supply rails; this has been the case ever since opamps became common in audio in the 1970s and it was established that electrolytic capacitors could work satisfactorily and reliably without DC bias on them. When low voltages are used, there is typically only one supply rail available, and it is necessary to bias the opamp outputs so they are halfway between the supply rail and ground. This is most economically done by an RC “V/2 generator” which supplies bias to all the stages. This is sometimes called a “half-rail generator”. There is no necessity that the bias point be exactly half of the supply rail. Some opamps clip earlier in one direction than the other, and a slight adjustment up or down of the half-rail may allow a larger symmetrical output swing. Having all the signal-carrying parts of the circuit displaced from ground means that DC blocking capacitors are frequently required, to keep DC out of volume controls and switches, counteracting any parts saving there may be in the power supply due to the single rail.

Table 5.1 Low-voltage opamps, ranked in order of voltage noise density
Device Supply voltage Voltage noise 1 kHz Current noise CMRR typ Slew rate Format Price ratio 100-off
AD8022 5–24 V 2.3 nV/√Hz 1 pA/√Hz –95 dB 50 V/μs Dual 1.64
LM4562 5–34 V 2.7 nV/√Hz 1.6 pA/√Hz –120 dB 20 V/μs Dual 0.67
AD8656 2.7–5.5 V 4 nV/√Hz No spec –100 dB 11 V/μs Single 1.00 ref
AD8397 3–24 V 4.5 nV/√Hz 1.5 pA/√Hz –90 dB 53 V/μs Dual 1.69
OPA2365 2.2–5.5 V 4.5 nV/√Hz 4 fA/√Hz –100 dB 25 V/μs Dual 1.05
AD8066 5–24 V 7 nV/√Hz 0.6 fA/√Hz –91 dB 180 V/μs Dual 1.73
AD8616 2.7–5 V 10 nV/√Hz 0.05 pA/√Hz –100 dB 12 V/μs Dual 1.00
AD826 5–36 V 15 nV/√Hz 1.5 pA/√Hz –80 dB 350 V/μs Dual 1.76
AD823 3–36 V 16 nV/√Hz 1 fA/√Hz –75 dB 22 V/μs Dual 2.73

Figure 5.1 shows series and shunt feedback stages biased by a common V/2 generator R10, R11, C10. The shunt stage has a voltage gain of 2.2 times, set by R2 and R3. Capacitor C1 performs DC-blocking at the input; R1 ensures that the DC level at the input cannot float up due to leakage through this capacitor. Likewise a blocking capacitor C2 is required to prevent DC flowing through the load.

Figure 5.1 Opamps running from a single supply rail, with a half-rail biasing generator shared between the two stages.

The series feedback stage has a voltage gain of 3.2 times, which means it works at the same noise gain as the shunt stage (see Chapters 1 and 22 for an explanation of noise gain), so comparative tests can be made in which the only difference is the common-mode voltage on the series stage opamp input pins. The series circuit requires an extra blocking capacitor C3 to prevent DC flowing through the feedback network and to reduce the gain to unity at zero frequency. An extra resistor R4 is needed to bias the stage input.

The V/2 generator typically uses a rather large capacitor to filter out disturbances on the supply rail. It is economical to share this with other stages, as shown in Figure 5.1. The low impedance of the filter capacitor should prevent interaction between stages if they are not of high gain, but isolation will tend to fall with frequency. If a good crosstalk performance at LF is required in stereo circuitry, it may be necessary to use separate V/2 generators for each channel.

One of the great advantages of dual-rail operation is that the opamp supply currents, which to some extent have the same half-wave rectified nature as those in a Class-B power amplifier, are inherently kept out of the signal ground. With single-rail operation, the negative supply pin of an opamp will be at 0 V, but any temptation to connect it to signal ground must be stoutly resisted, or the half-wave rectified currents flowing will cause serious distortion. It is essential to provide a separate power ground, as shown in Figure 5.1, which only joins the signal ground back at the power supply.

Opamps for 5 V Operation

Table 5.1 shows some candidates for low-voltage operation. The table has been restricted to those opamps capable of working from +5 V or lower. This therefore excludes the ubiquitous 5532 (with a minimum of 6 V), though it has been tested, with results given in the text. Those opamps that have been tested are shown in bold.

Prices are per package, not per opamp section. This puts the single AD8656 at a serious price disadvantage. Prices as at Jan 2020.

The NE5532 in +5 V Operation

When an attempt is made to run the 5532 from a +5 V rail, the maximum output (with no load beyond the shunt feedback network) is only 845 mVrms (at 1% THD) compared with the 1.77 Vrms theoretically available. This is only 48% of the possible voltage swing –less than half –and the 5532 output stage is clearly not performing well at a task it was never intended to do. Shunt feedback was used to exclude any common-mode distortion in the opamp input stage, with a gain of 2.2 times, as in Figure 5.1. Distortion is high, even at lower levels such as 700 mVrms, but falls below the relatively high noise level at 500 mV out. Perhaps most importantly, using any IC outside of its specifications is very risky, because one batch may work acceptably but another not at all.

The LM4562 in +5 V Operation

The much newer LM4562 opamp has a supply specification reaching lower down to 5 V, so we expect better low-voltage performance. We get it, too; the maximum output with no load is 1.16 Vrms (1% THD), a more reasonable 65% of the possible voltage swing, but this is still very inefficient compared with the same opamp working from ±17 V rails.

The distortion performance for three output levels, with shunt feedback and a gain of 2.2 times as in Figure 5.1, is shown in Figure 5.2. The flat traces simply represent noise, and no distortion is visible in the THD residual except around 10–20 kHz for a 1.0 Vrms output. You will note that the relative noise levels, and hence the minimum measurable THD levels, are considerably higher than for opamps on ±17 V rails.

Figure 5.2 LM4562 THD with +5 V supply for output levels of 0.5 V, 0.7 V, and 1.0 Vrms. No external load.

The LM4562 is thus a possibility for +5 V operation, but the 3.7 dB loss in maximum output caused by that 65% when the dynamic range is already being squeezed is very unwelcome. There is also the point that the opamp is working right on the lower limit of its supply specs, which is not ideal. I found some evidence that the noise performance was impaired compared with ±17 V rails, with a noticeable incidence of burst noise. The NE5532 is at least cheap, but the LM4562 is still relatively expensive. There may therefore be no economic disadvantage in using specialised low-voltage opamps which are also expensive but designed for the job. We will examine some of them now.

The AD8022 in +5 V Operation

The AD8022 from Analog devices looks like a possible candidate, having an exceptionally low input noise density of 2.3 nV/Hz that should allow maximisation of the dynamic range. It is a bipolar opamp with a wide supply range from +5 V to +24 V for single-rail operation, or ±2.5 V to ±12 V for dual rails; we would therefore still be operating at the lower limit of supply voltage. It is fabricated using a high-voltage bipolar process called XFCB.

First, the maximum output is 1.18 Vrms (no load, 1% THD), 67% of the available voltage swing; this is only a tiny improvement on the LM4562. The distortion performance, using shunt feedback and a gain of 2.2 times as in Figure 5.1, is not inspiring. According to the data sheet, we can expect second harmonic at -80 dB (0.01%) and third harmonic at -90 dB (0.003%) with a 0.7 Vrms output into a 500 Ω load, flat with frequency. Figure 5.3 shows the measured results for various output levels; they are markedly worse than the data sheet suggests, for reasons unknown. Note that we have changed the THD scale radically to accommodate the much higher levels, and the flat traces here really do represent distortion rather than noise. Several samples were tested with near-identical results.

Figure 5.3 AD8022 THD with +5 V supply for output levels of 0.3 V, 0.5 V, 0.7 V, and 1.0 Vrms. Shunt feedback, no external load. 80 kHz bandwidth.

These distortion levels are too high for any sort of quality audio, and the AD8022 will not be a good choice in most cases. No tests were therefore done on how distortion varies with output loading. However, for some applications, it may be worth bearing in mind that its input noise density is exceptionally low at 2.3 nV/Hz.

The AD8397 in +5 V Operation

The AD8397 is another promising possibility, because it claims a very voltage-efficient output stage and good load-driving capabilities. The input noise density is somewhat higher at 4.5 nV/Hz, slightly better than a 5532 and slightly worse than a 5534. It is a bipolar opamp with a wide supply range from 3 V to 24 V (for single-rail operation) or ±1.5 V to ±12 V for dual rails; this means that at +5 V, we are not operating at the lower supply voltage limit, which increases our confidence. It is fabricated using a complementary bipolar high voltage process called XFCB-HV.

The first good result is that the AD3897 really is voltage efficient. The maximum output with no external load is 1.83 Vrms at 1% THD. This is actually more than the theoretical output of 1.77 Vrms because the 1% THD criterion relies on some clipping to create the distortion. If we instead use as our output criterion the first hint of disturbance on the THD residual, we get exactly 1.77 Vrms, which is 100% voltage efficiency! Adding a 100 Ω load only reduces this to 1.72 Vrms.

The measured THD results, using shunt feedback and a gain of 2.2 times as in Figure 5.1, are shown in Figure 5.4 for various output levels. (Note that the noise gain is higher at 3.2x.) The flat parts of the traces represent noise only, with detectable distortion only appearing above 10 kHz, except at 1.76 Vrms out. The measured relative noise levels obviously fall as the output voltage increases.

The unloaded distortion results are very encouraging, but how are they affected by an external load on the output? Hardly at all, as Figure 5.5 shows; the output level is close to the maximum at 1.5 Vrms, and the external loads go down to 220 Ω, well below the minimum 5532 load of 500 Ω. Shunt feedback was used.

The AD8397 data sheet refers to loads down to an impressively low 25 Ω, so we want to explore the linearity with loading a bit further. Figure 5.6 shows the measured THD results with no load (NL) and external loads of 220 Ω, 150 Ω and 100 Ω, again at an output of 1.5 Vrms. Not until the load falls to 150 Ω do we start to see significant distortion appearing above 1 kHz, and this promises that this opamp will work well with low-impedance designs that will minimise noise and optimise the dynamic range.

So far, so good. The next thing to investigate is the distortion performance using series feedback, which puts a significant common-mode signal on the input pins. The test circuit used is the series stage in Figure 5.1, which has a gain of 3.2 times, but the same noise gain (3.2x) as the shunt circuit. Figure 5.7 shows the measured THD results with No Load (NL) and external loads of 220 Ω, 150 Ω and 100 Ω, again at an output of 1.5 Vrms. The distortion behaviour is clearly worse with series feedback for loads of 220 Ω or lower; care will be needed if low impedances are to be driven from a series feedback stage using an AD8397.

Figure 5.4 AD8397 THD with +5 V supply. Output levels of 0.5 V, 1.0 V, 1.5 V, and 1.76 Vrms. Shunt feedback, no external load.

Figure 5.5 AD8397 THD with +5 V supply. External loads from 680 Ω down to 220 Ω at an output level of 1.5 Vrms. Shunt feedback.

Figure 5.6 AD8397 THD with +5 V supply. With no load (NL) and external loads of 220 Ω, 150 Ω, and 100 Ω at an output level of 1.5 Vrms. Shunt feedback, gain =2.2x.

Figure 5.7 AD8397 THD with +5 V supply. With no load (NL) and external loads of 330 Ω, 220 Ω, 150 Ω, and 100 Ω at an output level of 1.5 Vrms. Series feedback, gain =3.2x.

The worst case for common-mode voltage is the voltage-follower configuration, where the full output voltage is present on the input terminals, and this exposes a limitation of the AD8397. The output stage does a magnificent job of using the whole available voltage swing, but the input stage does not quite have a full rail-to-rail capability. This means that voltage-follower stages, where the input and feedback signals are equal to the output signal, have a limitation on output level set by the input stage rather than the output stage. The distortion performance is also significantly worse than for the case of series feedback with a gain of 3.2x. Figure 5.8 demonstrates that distortion rises quickly as the signal level exceeds 0.9 Vrms. If you want to make use of the full output swing, then voltage-followers should be avoided. A relatively small amount of voltage gain, for example 1.1 times, will reduce the signal levels on the input pins and allow the full output swing with low distortion.

Figure 5.8 AD8397 voltage-follower THD with +5 V supply. No load (NL) and input/output levels of 0.7 V, 0.8 V, 0.9 V, 1.0 V, 1.05 V, and 1.1 Vrms.

Common mode distortion is worsened when the opamp is fed from a significant source resistance. Our final test is to drive a voltage-follower through a 4k7 resistance. Figure 5.9 shows that distortion is much increased compared with Figure 5.8. We had about 0.0025% THD with a 1.05 Vrms signal, but adding the 4k7 in series with the input has increased that to 0.007%.

Figure 5.9 AD8397 voltage-follower THD with 4k7 series input resistor. +5 V supply, no load (NL) and input/output levels of 0.7 V, 0.8 V, 0.9 V, 1.0 V, and 1.05 Vrms.

To conclude, the AD8397 is a useful opamp for +5 V operation; it is the best I have evaluated so far. It is, however, necessary to keep an eye on the effects of common-mode distortion and source resistance and be aware of the voltage limits of the input stage.

Opamps for 3.3 V Single-Rail Operation

First, forget about the NE5532. Its maximum output on a +3.3 V rail (with appropriate half-rail biasing) is only 240 mVrms (1% THD). This is a meagre 20% of the available voltage swing and is clearly very inefficient. There is considerable distortion and the residual shows nasty crossover spikes of the sort associated with under-biased power amplifiers. Finally, using any IC so far outside of its specifications is very dangerous indeed –one batch may work acceptably but another batch not at all.

The LM4562 works rather better from +3.3 V, giving 530 mVrms (1% THD). This is a slightly more respectable 45% of the available voltage swing, but the objection to using it outside of its specifications remains. The THD at output levels up to 430 mVrms is around 0.01%, which is not very encouraging.

Figure 5.10 AD8397 THD with +3 V3 supply. Output levels of 0.5 V, 0.7 V, 1.0 V, and 1.1 Vrms. Shunt feedback, no external load.

Figure 5.11 AD8397 THD with +3 V3 supply. With no load (NL) and external loads of 4k7, 2k2, 1k, 560 Ω, and 330 Ω at an output level of 1.1 Vrms. Shunt feedback, gain =2.2x.

Figure 5.12 AD8397 THD with +3 V3 supply. With external loads of 220 Ω, 150 Ω and 100 Ω at an output level of 1.1 Vrms. Shunt feedback, gain =2.2x.

The AD8397, however, works well at +3.3 V. The maximum output is now 1.18 Vrms (1% THD). The measured THD results, using shunt feedback and a gain of 2.2 times as in Figure 5.1, are shown in Figure 5.10 for differing output levels. The flat parts of the traces represent noise only, with detectable distortion only appearing above 10 kHz. Compare this with the +5 V results in Figure 5.4; there is little difference, though the maximum test level is of course less.

At +3.3 V, the AD8397 still works well with quite heavy loading. Figure 5.11 shows that loads lighter than 560 Ω have very little effect on the distortion performance, while 330 Ω causes detectable distortion only above 2 kHz. Figure 5.12 demonstrates that heavier loads than this lead to significant distortion. The AD8022 was not evaluated for +3.3 V operation, as its minimum supply voltage is +5 V.