Back to the basics: Achieve low distortion and noise audio design with Class AB amps - Embedded.com

Back to the basics: Achieve low distortion and noise audio design with Class AB amps

Audio design has always stirred the passions of most electronicsengineers. It pursues perfection before cost considerations. Theseaside, sometimes, the simplest circuits can offer the best solutionsfor audio design in terms of cost and performance. A quick look at manynew low-power speaker amplifiers on the market highlights the move toclass D audio performance.

But when it comes to low cost, distortion and noise, class AB stillhas the edge. The class AB architecture offers a signal-to-noise plusdistortion ratio of up to 10 times better than its equivalent class Dneighbor.

It provides a much simpler architecture without the need forreactive filter components on the output and the electromagneticradiation resulting from an output stage switching at a few hundredkilohertz. However, a class AB output stage is only part of thearchitecture of a low noise, low-cost audio amplifier.

System-level design issues
Consideration of the system-level design reveals that good audioperformance at low frequencies requires the use of large couplingcapacitors to the speaker. Even with a relatively high-impedancespeaker, a capacitor of a few thousand microfarads is needed to providesufficiently low impedance to provide enough drive.

A differential class AB output stage solves this problem by enablingthe speaker to be driven directly without the need for a DC blockingcapacitor. In addition, if each side of the differential output swingsin antiphase, the load sees twice the voltage output compared withconventional single-ended drive.

A differential input stage also yields benefits with the noiseperformance of the amplifier. Noise can contaminate a signal in twoways: noise on the ground line and noise coupled onto the signal. Inall mixed-signal audio designs, there will be some digital circuitryclose by, and often, the ideal of laying out two ground planes (oneeach for analog and digital) is impossible to achieve.

Hence, high-speed digital noise is often coupled onto the analogground, which can reduce any signal tonoise (SNR) advantages that a high-performance amplifier mayboast. In high-precision analog design, it is always safe to assumethat the ground rail will never be truly at 0V.

There will be low-frequency noise often caused by currents flowinginto and out of the power pins of analog components. There will also behigh-frequency noise often caused by digital circuitry taking sharpsurges of current into their power pins and coupling noise onto theground from the signal lines. If an input signal is referenced withrespect to a noisy ground, this noise will be faithfully reproduced atthe output of the amplifier.

In addition to noise on the ground line of the amplifier, noise canalso be coupled onto the signal from surrounding circuitry. Inside ahandset, for example, the signal may be subjected to noise from the RFcircuitry, high-speed digital circuitry and noise from severalswitching regulators.

Overcoming ground and signal noise
A differential input stage overcomes both ground and signal noises, andallows only the signal of interest to be amplified. If the PCB tracescarrying the input signal are routed closely and parallel to eachother, any noise coupled onto the input will be common to both inputsand hence, not amplified.

Moreover, in a well-designed differential input stage, the closematching of the input circuitry means that any distortion generated ateach input is normally equal on both inputs and cancelled out by thedifferential nature of the input. Figure1 below shows a 1.8W class AB differential input, differentialoutput audio amplifier from austriamicrosystems.

Figure1: To analyze the circuit, consider the currents flowing into theinverting and non-inverting differential inputs.

To analyze the circuit, consider the currents flowing into theinverting and non-inverting differential inputs. Since the amplifierinput impedance is high, any current flowing into RIN1 flows throughRF1 and into the output. The same happens with RIN2 and RF2. Thus,

[Vin(-) – V(-)] / RIN1 = [{V(-) -Vout(+)}xV]/RF1

when the amplifier is in its linear mode V(-) = V(+). If RF1 = RF2 =RF and RIN1 = RIN2 = RIN, equating V(+) and V(-)eventually gives:

Vout(-) – Vout(+) = [RF/RIN]/[Vin(-) – Vin(+)]

In other words, the differential signal on the output is equal tothe differential signal on the input, multiplied by the ratio of thefeedback resistors to input resistors – no different from aconventional op amp.

A real-life circuit: bass-boostamplifier
No application note would be complete without a real-life circuit. Thecircuit chosen was a bass-boost amplifier to take the output from aconventional MP3 chipset (in this case, the AS3525 single-chip MP3 player) andamplify its frequency response below 100Hz to provide some bass liftand decent drive to power a PC loudspeaker.

Many people have MP3 players and PCs with speakers, so it wasprudent to design a circuit that merges the two together, enabling theuser to listen to MP3 audio through existing PC speakers. The amplifierwas designed around the AS1702 1.8W audio amplifier. Figure 2 below shows the final circuit diagram.

Figure2: The feedback resistor, RF1, is bypassed by the frequency-dependentcircuit comprising Cs and Rs.

The feedback resistor, RF1, is bypassed by the frequency dependentcircuit comprising Cs and Rs. At low frequencies, Cs is high-impedanceand the gain is simply:

RF1/RIN = 300/75 = 4

As the frequency increases, Cs provides a lowering impedance inparallel with RF1, thus reducing the gain.

The series resistor, Rs, provides a zero to level off the decreasinggain and provide unity gain at high frequencies. The same feedbacknetwork is repeated for the other differential input. The gain of thecircuit is proportional to:

EQ1

Equating the denominator to zero determines the start of theroll-off, and equating the numerator to zero determines the levelingoff frequency. Choosing components of Cs = 5nF, Rs = 100k ohms , RF1 =300k ohms and RIN = 75k ohms yields a roll-off starting at about 80Hzand leveling off at about 318Hz.

Figure3a: Equating the denominator to zero determines the start of therolloff, and the numerator to zero determines the leveling offfrequency.

Figure 3a above shows thetheoretical roll-off characteristic (although Excel does not computephase), and Figure 3b below shows a plot of the practical results taken. Practical listening testsof this circuit proved that it did boost the bass and provide goodperformance. To give a wider bandwidth of bass boost, changing theseries capacitor from 5nF to 2nF takes the roll-off frequency out to200Hz.

Figure3b: Practical listening tests of this circuit proved that it did boostthe bass and provide good performance.

The circuit was built using standard 2 percent tolerance resistorsand 10 percent tolerance capacitors. The distortion was then measuredusing HP's HP339Adistortionmeter.The distortion of the bass boost amplifier (Figure 4 below ) shows that this isacceptable for many handheld audio applications considering thecomponents used.

Figure4: The distortion of the bass boost amplifier shows that this isacceptable for many handheld audio applications considering thecomponents used.

Given that distortion is directly related to the matching of thefeedback components, it is important to keep these as accurate as thedesign budget allows.

An interesting experiment
As an experiment, it was interesting to see the effect of shorting outthe series feedback capacitor Cs, since this had a large tolerance andmay contribute significantly to the distortion (Figure 5 below ). Indeed, thecapacitor contributed to the distortion above the rolloff frequency.However, in both cases, the distortion remained below 0.7 percent formuch of the audio band.

Figure5: As an experiment, it was interesting to see the effect of shortingout the series feedback capacitor Cs, since this had a large toleranceand may contribute significantly to the distortion.

It was then decided to take some spot measurements with a precisionresistor network around the AS1702. A Vishay ORNA2-1 resistor network was used comprising two 10k ohm and two5k ohm resistors in one package, both trimmed to a ratiometrictolerance of 0.05 percent. The results did not differ significantlyfrom those plotted in Figure 5, above.

This indicates that the distortion found in Figure 5 was largely dueto the internal distortion of the AS1702, and not from the surroundingresistor network (the series capacitor at this time was stillshort-circuited).

To further prove that the two feedback resistors and two inputresistors need to keep a close tolerance of each other, a 680 kohmresistor was placed in parallel with one 300 kohms  feedbackresistor, RF1.

Then, spot distortion measurements were taken. Distortion at 10Hzwas 1.5 percent, at 30Hz it was 1.35 percent, and from 100Hz to 1kHz itwas 1.2 percent. Comparing this with the 0.06 percent distortion shownin Figure 5 gives practicalbackup to the theory that the feedbackresistors must be identical to gain optimum performance.

Conclusion
The differential input/differential output amplifier is similar to theconventional single-ended output op amp in terms of the equations thatdetermine its gain. Results show that the two input resistors and twofeedback resistors must be similar to achieve optimum performance.Looking at the effect of using 2 percent tolerance resistors, however,shows that the circuit is quite forgiving when measuring the resultingdistortion.

Simon Bramble is North EuropeanApplications Engineer at austriamicrosystems. 

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.